KR102421718B1 - A method and an apparatus for performing artificial intelligence encoding and artificial intelligence decoding - Google Patents

Abstract

a memory storing one or more instructions; and a processor executing the one or more instructions stored in the memory, wherein the processor determines the size of the resolution of the original image, and if the resolution of the original image is greater than a predetermined value, the downscaling DNN Acquires an AI downscaled first image from an original image, and if the resolution of the original image is less than or equal to the predetermined value, AI 1 from the original image through DNN for one-to-one pre-processing for upscaling Obtaining a one-to-one preprocessed first image, generating image data by first encoding the first image, and AI data including information related to the AI downscale or information related to the AI one-to-one preprocessing, and the transmit image data, wherein the AI data includes information for selection of DNN setting information of a DNN for upscaling for AI upscaling of a second image, wherein the second image includes first decoding of the image data generated through, and the DNN setting information of the DNN for downscaling is obtained through a first joint training of the DNN for downscaling and the DNN for upscaling for AI upscaling of the second image, and the one-to-one The DNN setting information of the DNN for preprocessing uses the DNN setting information of the DNN for upscaling obtained through the first linked training, and the second linked training of the DNN for the one-to-one preprocessing and the DNN for the upscaling. We propose an AI encoding apparatus for an image, characterized in that it is obtained through

Description

A method and an apparatus for performing artificial intelligence encoding and artificial intelligence decoding

The present disclosure relates to a method and apparatus for processing an image, and more particularly, the present disclosure relates to a method and apparatus for performing artificial intelligence (AI) encoding and artificial intelligence decoding.

The image is encoded by a codec conforming to a predetermined data compression standard, for example, a Moving Picture Expert Group (MPEG) standard, and then stored in a recording medium in the form of a bitstream or transmitted through a communication channel.

With the development and dissemination of hardware capable of reproducing and storing high-resolution/high-definition images, the need for a codec capable of effectively encoding and decoding high-resolution/high-definition images is increasing.

In addition, the need for a codec capable of encoding a low-resolution image and decoding the transmitted low-resolution image into a high-resolution/high-quality image is also increasing.

The AI encoding and AI decoding method and apparatus of an image according to an embodiment have a technical task of encoding and decoding an image based on AI to achieve a low bit rate.

In order to solve the above technical problem, a video encoding apparatus proposed by the present disclosure includes: a memory for storing one or more instructions; and a processor executing the one or more instructions stored in the memory, wherein the processor determines the size of the resolution of the original image, and if the resolution of the original image is greater than a predetermined value, the original image through DNN for downscaling Acquires an AI downscaled first image from an image, and if the resolution of the original image is less than or equal to the predetermined value, one AI from the original image through DNN for one-to-one pre-processing for upscaling 1 Obtaining a first image preprocessed, generating image data by first encoding the first image, AI data including information related to the AI downscale or information related to the AI one-to-one preprocessing, and the image data is transmitted, and the AI data includes information for selection of DNN setting information of a DNN for upscaling for AI upscaling of a second image, and the second image performs the first decoding of the image data. generated through, and the DNN setting information of the downscaling DNN is obtained through the first joint training of the downscaling DNN and the upscaling DNN for the AI upscaling of the second image, and the one-to-one preprocessing The DNN setting information of the DNN for DNN is, by using the DNN setting information of the DNN for upscaling obtained through the first linked training, through the second linked training of the DNN for the one-to-one preprocessing and the DNN for the upscaling. may have been obtained.

In order to solve the above technical problem, a video encoding method proposed by the present disclosure includes: determining a resolution of an original image; if the resolution of the original image is less than or equal to a predetermined value, acquiring a first image pre-processed by AI through a DNN for one-to-one preprocessing; if the resolution of the original image is greater than the predetermined value, obtaining an AI downscaled first image from the original image through a downscaling DNN; generating image data by first encoding the first image; and transmitting AI data and the image data including information related to the AI one-to-one pre-processing or information related to the AI downscaling, wherein the AI data is upscaled for AI upscaling of the second image. information for selection of DNN configuration information of the DNN for scale, the second image is generated through the first decoding of the image data, and the DNN configuration information of the DNN for downscaling is the downscale DNN and DNN setting information of the DNN for upscaling for AI upscaling of the second image is obtained through a first joint training, and the DNN setting information of the DNN for one-to-one preprocessing is the upscaling obtained through the first joint training It may be obtained through the second joint training of the one-to-one preprocessing DNN and the upscaling DNN using the DNN setting information of the DNN for DNN.

In order to solve the above technical problem, a video decoding apparatus proposed by the present disclosure includes: a memory for storing one or more instructions; and a processor that executes the one or more instructions stored in the memory, wherein the processor includes image data generated as a result of a first encoding of a first image and AI downscale from an original image to the first image, or one AI 1 Acquire AI data related to preprocessing, first decode the image data to obtain a second image corresponding to the first image, and set a DNN for AI upscaling of the second image among a plurality of DNN setting information Obtaining information based on the AI data, and generating an AI upscaled third image from the second image through a Deep Neural Network (DNN) for upscaling that operates with the obtained DNN setting information, The DNN setting information of the DNN setting information for the upscaling DNN and the downscaling DNN used for the AI downscaling of the original image, and the DNN setting information for the AI upscaling obtained as a result of the first cooperative training It may be obtained through second joint training of the DNN for one-to-one pre-processing and the DNN for upscaling used for AI one-to-one pre-processing of the original image.

In order to solve the above technical problem, the video decoding method proposed by the present disclosure relates to image data generated as a result of the first encoding of a first image and AI downscaling from the original image to the first image or AI one-to-one preprocessing. acquiring AI data; obtaining a second image corresponding to the first image by first decoding the image data; obtaining DNN setting information for AI upscaling of the second image from among a plurality of DNN setting information, based on the AI data; and generating an AI upscaled third image from the second image through a deep neural network (DNN) for upscaling that operates with the obtained DNN setting information, wherein the plurality of DNN setting information includes: Using the first coordinated training of the DNN for upscaling and the DNN for downscaling used for the AI downscaling of the original image, and the DNN setting information for AI upscaling obtained as a result of the first coordinated training, the The one-to-one pre-processing DNN used for the AI one-to-one pre-processing may be obtained through the second linkage training of the upscaling DNN.

The resolution of the input image is determined, the downscaled image of the high-resolution image is transmitted, and the upscaled image is restored, and the low-resolution image is a one-to-one pre-processed image that preserves the detailed information of the image. By transmitting and upscaling and restoring, a high-resolution image based on AI transmits the downscaled image and then upscales it, efficiently transmitting the image and restoring it to a high-resolution image, and the content provider of the low-resolution image Without creating a new image, by preserving and transmitting detailed information of an existing low-resolution image and then upscaling, a high-resolution image with improved image quality can be generated by preventing image quality deterioration.

In order to more fully understand the drawings cited herein, a brief description of each drawing is provided.
1 is a diagram for explaining an AI encoding process and an AI decoding process according to an embodiment.
2 is a block diagram illustrating a configuration of an AI decoding apparatus according to an embodiment.
3 is an exemplary diagram illustrating a second DNN for AI upscaling of a second image.
4 is a diagram for explaining a convolution operation using a convolution layer.
5 is an exemplary diagram illustrating a mapping relationship between various pieces of image-related information and various pieces of DNN configuration information.
6 is a diagram illustrating a second image composed of a plurality of frames.
7 is a block diagram illustrating a configuration of an AI encoding apparatus according to an embodiment.
8 is an exemplary diagram illustrating a first DNN for AI downscaling of an original video.
9 is a diagram illustrating a structure of AI encoded data according to an embodiment.
10 is a diagram illustrating a structure of AI encoded data according to another embodiment.
11 is a diagram for explaining a method of training a first DNN and a second DNN.
12 is a diagram for explaining a training process of a first DNN and a second DNN by a training apparatus.
13 is a diagram for describing an artificial intelligence (AI) encoding process and an AI decoding process according to an embodiment.
14 is an exemplary diagram illustrating a third DNN for AI one-to-one pre-processing of an original image.
15 is a block diagram illustrating a configuration of an AI encoding apparatus according to an embodiment.
16 is a diagram for explaining a method of training a third DNN in association with a second DNN.
17 is a diagram for explaining a training process of a third DNN and a second DNN by a training apparatus.
18 is a flowchart illustrating an AI decoding method according to an embodiment.
19 is a flowchart illustrating an AI encoding method according to an embodiment.
20 is a diagram for explaining an artificial intelligence (AI) encoding process and an AI decoding process according to an embodiment.
21 is a block diagram illustrating a configuration of an AI encoding apparatus according to an embodiment.

Since the present disclosure can make various changes and can have various embodiments, specific embodiments are illustrated in the drawings and will be described in detail through the detailed description. However, this is not intended to limit the embodiments of the present disclosure, and it should be understood that the present disclosure includes all modifications, equivalents and substitutes included in the spirit and scope of various embodiments.

In describing the embodiment, if it is determined that a detailed description of a related known technology may unnecessarily obscure the subject matter of the present disclosure, the detailed description thereof will be omitted. In addition, numbers (eg, first, second, etc.) used in the description process of the specification are only identifiers for distinguishing one component from other components.

In addition, in this specification, when a component is referred to as "connected" or "connected" with another component, the component may be directly connected or directly connected to the other component, but in particular It should be understood that, unless there is a description to the contrary, it may be connected or connected through another element in the middle.

In addition, in the present specification, components expressed as '~ part (unit)', 'module', etc. are two or more components combined into one component, or two or more components for each more subdivided function. may be differentiated into In addition, each of the components to be described below may additionally perform some or all of the functions of other components in addition to the main functions that each component is responsible for, and some of the main functions of each of the components may have different functions. It goes without saying that it may be performed exclusively by the component.

Also, in this specification, an 'image' or 'picture' may indicate a still image, a moving picture composed of a plurality of continuous still images (or frames), or a video.

Also, in the present specification, a 'deep neural network (DNN)' is a representative example of an artificial neural network model simulating a brain nerve, and is not limited to an artificial neural network model using a specific algorithm.

Also, in the present specification, a 'parameter' is a value used in a calculation process of each layer constituting a neural network, and may include, for example, a weight used when an input value is applied to a predetermined calculation expression. Also, the parameter may be expressed in a matrix form. A parameter is a value set as a result of training, and may be updated through separate training data if necessary.

In addition, in this specification, 'first DNN' means a DNN used for AI downscaling of an image, and 'second DNN' means a DNN used for AI upscaling of an image.

In addition, in the present specification, 'DNN configuration information' includes the aforementioned parameters as information related to elements constituting the DNN. The first DNN or the second DNN may be configured using the DNN configuration information.

In addition, in this specification, the 'original image' refers to an image to be subjected to AI encoding, and the 'first image' refers to an image obtained as a result of AI downscaling of the original image in the AI encoding process. In addition, the 'second image' refers to an image obtained through the first decoding in the AI decoding process, and the 'third image' refers to an image obtained by AI upscaling the second image in the AI decoding process.

In addition, in this specification, 'AI downscaling' refers to a process of reducing the resolution of an image based on AI, and 'first encoding' refers to an encoding process by a frequency transform-based image compression method. In addition, 'first decoding' refers to a decoding process by a frequency transformation-based image restoration method, and 'AI upscaling' refers to a process of increasing the resolution of an image based on AI.

1 is a diagram for explaining an artificial intelligence (AI) encoding process and an AI decoding process according to an embodiment.

As described above, as the resolution of an image rapidly increases, the amount of information processing for encoding/decoding increases. Accordingly, a method for improving encoding and decoding efficiency of an image is required.

As shown in FIG. 1 , according to an embodiment of the present disclosure, a first image 115 is acquired by AI downscaling 110 of an original image 105 having a high resolution. In addition, since the first encoding 120 and the first decoding 130 are performed on the first image 115 of a relatively small resolution, the first encoding 120 and Compared to the case of performing the first decoding 130 , the bit rate may be greatly reduced.

1 , in an embodiment, in the AI encoding process, the original image 105 is AI downscaled 110 to obtain a first image 115 , and the first image 115 is first Encoding (120). In the AI decoding process, AI encoding data including AI data and image data obtained as a result of AI encoding is received, a second image 135 is obtained through the first decoding 130, and the second image 135 is A third image 145 is obtained by AI upscaling 140 .

Looking at the AI encoding process in more detail, when the original image 105 is input, the original image 105 is AI downscaled 110 to obtain the first image 115 of a predetermined resolution and/or a predetermined image quality. At this time, the AI downscaling 110 is performed based on AI, the AI for the AI downscaling 110 must be trained in connection with the AI for the AI upscaling 140 of the second image 135 (joint trained) do. Because, when the AI for the AI downscaling 110 and the AI for the AI upscaling 140 are separately trained, between the original image 105 that is the AI encoding target and the third image 145 restored through AI decoding because the difference between

In an embodiment of the present disclosure, AI data may be used in order to maintain this linkage in the AI encoding process and the AI decoding process. Therefore, the AI data obtained through the AI encoding process should include information indicating the upscale target, and in the AI decoding process, the second image 135 is AI upscaled ( 140) should be done.

AI for AI downscaling 110 and AI for AI upscaling 140 may be implemented as a deep neural network (DNN). As will be described later with reference to FIG. 11 , since the first DNN and the second DNN are jointly trained through sharing of loss information under a predetermined target, the AI encoding apparatus uses target information used when the first DNN and the second DNN are jointly trained. is provided to the AI decoding apparatus, and the AI decoding apparatus may upscale the AI to the image quality and/or resolution targeting the second image 135 based on the received target information.

When the first encoding 120 and the first decoding 130 shown in FIG. 1 are described in detail, the first image 115 AI downscaled 110 from the original image 105 is the first encoding 120 . can reduce the amount of information. The first encoding 120 includes a process of predicting the first image 115 to generate prediction data, a process of generating residual data corresponding to a difference between the first image 115 and the prediction data, and a spatial domain component. It may include a process of transforming the residual data into a frequency domain component, a process of quantizing the residual data transformed into a frequency domain component, and a process of entropy encoding the quantized residual data. Such a first encoding process 120 is MPEG-2, H.264 Advanced Video Coding (AVC), MPEG-4, High Efficiency Video Coding (HEVC), VC-1, VP8, VP9 and AV1 (AOMedia Video 1). It can be implemented through one of the image compression methods using frequency transformation, etc.

The second image 135 corresponding to the first image 115 may be restored through the first decoding 130 of image data. The first decoding 130 includes a process of generating quantized residual data by entropy-decoding image data, a process of inverse quantizing the quantized residual data, a process of transforming the residual data of a frequency domain component into a spatial domain component, a process of predicting data It may include a process of generating , and a process of reconstructing the second image 135 using prediction data and residual data. The first decoding 130 process as described above is an image compression using frequency conversion such as MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9 and AV1 used in the first encoding 120 process. It may be implemented through an image restoration method corresponding to one of the methods.

The AI encoded data obtained through the AI encoding process may include image data obtained as a result of the first encoding 120 of the first image 115 and AI data related to the AI downscale 110 of the original image 105. can The image data may be used in the first decoding 130 process, and the AI data may be used in the AI upscaling 140 process.

The image data may be transmitted in the form of a bitstream. The image data may include data obtained based on pixel values in the first image 115 , for example, residual data that is a difference between the first image 115 and prediction data of the first image 115 . . Also, the image data includes information used in the process of the first encoding 120 of the first image 115 . For example, the image data includes prediction mode information used for the first encoding 120 of the first image 115 , motion information, and quantization parameter related information used in the first encoding 120 , etc. can do. The image data is the image compression method used in the first encoding 120 process among the image compression methods using frequency conversion such as MPEG-2, H.264 AVC, MPEG-4, HEVC, VC-1, VP8, VP9 and AV1. It may be generated according to a rule, for example, a syntax.

AI data is used for AI upscaling 140 based on the second DNN. As described above, since the first DNN and the second DNN are jointly trained, the AI data includes information enabling the accurate AI upscaling 140 of the second image 135 through the second DNN to be performed. . In the AI decoding process, the AI upscaling 140 may be performed to a resolution and/or image quality targeting the second image 135 based on the AI data.

AI data may be transmitted together with image data in the form of a bitstream. According to an embodiment, AI data may be transmitted separately from image data in the form of frames or packets.

Alternatively, according to an embodiment, AI data may be transmitted while being included in image data.

The image data and AI data may be transmitted through the same network or different networks.

2 is a block diagram illustrating a configuration of an AI decoding apparatus 200 according to an embodiment.

Referring to FIG. 2 , the AI decoding apparatus 200 according to an embodiment includes a receiving unit 210 and an AI decoding unit 230 . The AI decoding unit 230 may include a parsing unit 232 , a first decoding unit 234 , an AI upscaling unit 236 , and an AI setting unit 238 .

Although the receiver 210 and the AI decoder 230 are illustrated as separate devices in FIG. 2 , the receiver 210 and the AI decoder 230 may be implemented through one processor. In this case, the receiving unit 210 and the AI decoding unit 230 may be implemented as a dedicated processor, and a general-purpose processor such as an application processor (AP), a central processing unit (CPU), or a graphic processing unit (GPU) and S/W It may be implemented through a combination of In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

The receiver 210 and the AI decoder 230 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU. In one embodiment, the receiving unit 210 is implemented as a first processor, the first decoding unit 234 is implemented as a second processor different from the first processor, the parsing unit 232, the AI upscaling unit 236 and the AI setting unit 238 may be implemented as a third processor different from the first processor and the second processor.

The receiving unit 210 receives AI encoded data obtained as a result of AI encoding. As an example, the AI-encoded data may be a video file having a file format such as mp4 or mov.

The receiver 210 may receive AI-encoded data transmitted through a network. The receiving unit 210 outputs the AI encoded data to the AI decoding unit 230 .

In one embodiment, the AI-encoded data is a hard disk, magnetic media such as floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks. It may be obtained from a data storage medium including an optical medium).

The parsing unit 232 parses the AI encoded data and transmits image data generated as a result of the first encoding of the first image 115 to the first decoder 234 , and transmits the AI data to the AI setting unit 238 . transmit

In an embodiment, the parsing unit 232 may parse the image data and the AI data separately included in the AI encoded data. The parsing unit 232 may read the header in the AI-encoded data to distinguish the AI data and the image data included in the AI-encoded data. In one example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream.

The structure of AI encoded data including AI data and image data separated from each other will be described later with reference to FIG. 9 .

In another embodiment, the parsing unit 232 parses the image data from the AI encoded data, extracts the AI data from the image data, transmits the AI data to the AI setting unit 238, and first decodes the remaining image data. may be transmitted to the unit 234 . That is, AI data may be included in image data. For example, AI data may be included in supplemental enhancement information (SEI), which is an additional information area of a bitstream corresponding to image data. The structure of AI-encoded data including image data including AI data will be described later with reference to FIG. 10 .

In another embodiment, the parsing unit 232 divides the bitstream corresponding to the image data into a bitstream to be processed by the first decoding unit 234 and a bitstream corresponding to AI data, and divides each divided bitstream It may output to the first decoding unit 234 and the AI setting unit 238 .

The parsing unit 232 obtains image data included in the AI encoded data through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1). It can also be confirmed that it is the image data that has been processed. In this case, the corresponding information may be transmitted to the first decoder 234 so that the image data can be processed by the identified codec.

The first decoder 234 reconstructs the second image 135 corresponding to the first image 115 based on the image data received from the parser 232 . The second image 135 obtained by the first decoder 234 is provided to the AI upscaler 236 .

According to an embodiment, first decoding-related information such as prediction mode information, motion information, and quantization parameter information may be provided from the first decoding unit 234 to the AI setting unit 238 . The first decoding related information may be used to obtain DNN configuration information.

The AI data provided to the AI setting unit 238 includes information enabling AI upscaling of the second image 135 . In this case, the upscale target of the second image 135 should correspond to the downscale target of the first DNN. Therefore, the AI data should include information that can confirm the downscale target of the first DNN.

Specifically exemplifying the information included in the AI data, there are information about a difference between the resolution of the original image 105 and the resolution of the first image 115 , and information related to the first image 115 .

The difference information may be expressed as information about a resolution conversion degree of the first image 115 compared to the original image 105 (eg, resolution conversion rate information). And, since the resolution of the first image 115 is known through the resolution of the restored second image 135 and the degree of resolution conversion can be confirmed through this, the difference information can be expressed only with the resolution information of the original image 105 . may be Here, the resolution information may be expressed as a horizontal/vertical screen size, or it may be expressed as a ratio (16:9, 4:3, etc.) and a size of one axis. In addition, when there is preset resolution information, it may be expressed in the form of an index or a flag.

The information related to the first image 115 includes the resolution of the first image 115 , the bit rate of image data obtained as a result of the first encoding of the first image 115 , and the first encoding of the first image 115 . It may include information on at least one of the codec types used at the time of operation.

The AI setting unit 238 may determine an upscale target of the second image 135 based on at least one of difference information included in the AI data and information related to the first image 115 . The upscaling target may indicate, for example, to which resolution the second image 135 should be upscaled. When the upscaling target is determined, the AI upscaling unit 236 AI upscales the second image 135 through the second DNN to obtain a third image 145 corresponding to the upscaling target.

Prior to explaining a method for the AI setting unit 238 to determine an upscaling target based on AI data, an AI upscaling process through the second DNN will be described with reference to FIGS. 3 and 4 .

3 is an exemplary diagram illustrating a second DNN 300 for AI upscaling of a second image 135, and FIG. 4 is a convolution operation in the first convolution layer 310 shown in FIG. is showing

3 , the second image 135 is input to the first convolutional layer 310 . 3X3X4 displayed in the first convolution layer 310 shown in FIG. 3 exemplifies convolution processing on one input image using four filter kernels having a size of 3×3. As a result of the convolution process, four feature maps are generated by four filter kernels. Each feature map represents unique characteristics of the second image 135 . For example, each feature map may represent a vertical direction characteristic, a horizontal direction characteristic, or an edge characteristic of the second image 135 .

A convolution operation in the first convolution layer 310 will be described in detail with reference to FIG. 4 .

Through the multiplication operation and addition operation between the parameters of the filter kernel 430 having a size of 3 X 3 used in the first convolution layer 310 and the pixel values in the second image 135 corresponding thereto, one A feature map 450 may be generated. Since four filter kernels are used in the first convolution layer 310 , four feature maps may be generated through a convolution operation process using the four filter kernels.

In FIG. 4 , I1 to I49 displayed on the second image 135 indicate pixels of the second image 135 , and F1 to F9 displayed on the filter kernel 430 indicate parameters of the filter kernel 430 . Also, M1 to M9 displayed in the feature map 450 indicate samples of the feature map 450 .

4 illustrates that the second image 135 includes 49 pixels, but this is only an example, and when the second image 135 has a resolution of 4K, for example, 3840 X 2160 pixels It may contain pixels.

In the convolution operation process, pixel values of I1, I2, I3, I8, I9, I10, I15, I16, and I17 of the second image 135 and F1, F2, F3, F4, and F5 of the filter kernel 430 are respectively , F6, F7, F8, and F9 may each be multiplied, and a value obtained by combining (eg, addition operation) result values of the multiplication operation may be assigned as the value of M1 of the feature map 450 . If the stride of the convolution operation is 2, pixel values of I3, I4, I5, I10, I11, I12, I17, I18, and I19 of the second image 135 and F1 and F2 of the filter kernel 430, respectively , F3, F4, F5, F6, F7, F8, and F9 may each be multiplied, and a value obtained by combining result values of the multiplication operation may be assigned as the value of M2 of the feature map 450 .

A convolution operation between the pixel values in the second image 135 and the parameters of the filter kernel 430 is performed while the filter kernel 430 moves along the stride until the last pixel of the second image 135 is reached. By being performed, a feature map 450 having a predetermined size may be obtained.

According to the present disclosure, parameters of the second DNN through joint training of the first DNN and the second DNN, for example, parameters of the filter kernel used in convolutional layers of the second DNN (eg, filter The values of F1, F2, F3, F4, F5, F6, F7, F8, and F9 of the kernel 430 may be optimized. The AI setting unit 238 determines an upscale target corresponding to the downscale target of the first DNN based on the AI data, and uses parameters corresponding to the determined upscale target in the convolution layers of the second DNN. It can be determined by parameters of the filter kernel.

The convolution layers included in the first DNN and the second DNN may be processed according to the convolution operation process described with reference to FIG. 4 , but the convolution operation process described in FIG. 4 is only an example, and the it is not

Referring again to FIG. 3 , the feature maps output from the first convolutional layer 310 are input to the first activation layer 320 .

The first activation layer 320 may provide a non-linear characteristic to each feature map. The first activation layer 320 may include a sigmoid function, a Tanh function, a Rectified Linear Unit (ReLU) function, and the like, but is not limited thereto.

Giving the nonlinear characteristic in the first activation layer 320 means changing and outputting some sample values of the feature map, which is an output of the first convolution layer 310 . At this time, the change is performed by applying a non-linear characteristic.

The first activation layer 320 determines whether to transfer sample values of the feature maps output from the first convolution layer 310 to the second convolution layer 330 . For example, some sample values among the sample values of the feature maps are activated by the first activation layer 320 and transmitted to the second convolution layer 330 , and some sample values are activated by the first activation layer 320 . It is deactivated and is not transmitted to the second convolutional layer 330 . The intrinsic characteristic of the second image 135 indicated by the feature maps is emphasized by the first activation layer 320 .

The feature maps 325 output from the first activation layer 320 are input to the second convolution layer 330 . Any one of the feature maps 325 shown in FIG. 3 is a result of processing the feature map 450 described with reference to FIG. 4 in the first activation layer 320 .

3X3X4 displayed in the second convolution layer 330 exemplifies convolution processing on the input feature maps 325 using four filter kernels having a size of 3×3. The output of the second convolutional layer 330 is input to the second activation layer 340 . The second activation layer 340 may provide a non-linear characteristic to the input data.

The feature maps 345 output from the second activation layer 340 are input to the third convolution layer 350 . 3X3X1 displayed in the third convolution layer 350 shown in FIG. 3 exemplifies convolution processing to generate one output image using one filter kernel having a size of 3×3. The third convolution layer 350 is a layer for outputting a final image and generates one output using one filter kernel. According to the example of the present disclosure, the third convolution layer 350 may output the third image 145 through a convolution operation.

DNN setting information indicating the number of filter kernels of the first convolutional layer 310, the second convolutional layer 330, and the third convolutional layer 350 of the second DNN 300, parameters of the filter kernel, etc. As will be described later, there may be a plurality of pieces, and a plurality of pieces of DNN configuration information should be associated with a plurality of pieces of DNN configuration information of the first DNN. The association between the plurality of DNN configuration information of the second DNN and the plurality of DNN configuration information of the first DNN may be implemented through association learning of the first DNN and the second DNN.

3 shows that the second DNN 300 includes three convolutional layers 310 , 330 , 350 and two activation layers 320 , 340 , but this is only an example, and the implementation Accordingly, the number of convolutional layers and activation layers may be variously changed. Also, according to an embodiment, the second DNN 300 may be implemented through a recurrent neural network (RNN). In this case, it means changing the CNN structure of the second DNN 300 according to the example of the present disclosure to the RNN structure.

In an embodiment, the AI upscaling unit 236 may include at least one arithmetic logic unit (ALU) for the above-described convolution operation and operation of the activation layer. The ALU may be implemented as a processor. For the convolution operation, the ALU includes a multiplier that performs a multiplication operation between the sample values of the feature map output from the second image 135 or the previous layer and the sample values of the filter kernel, and an adder that adds the result values of the multiplication. can In addition, for the calculation of the activation layer, the ALU is a multiplier that multiplies an input sample value by a weight used in a predetermined sigmoid function, a Tanh function, or a ReLU function, and compares the multiplication result with a predetermined value to calculate the input sample value It may include a comparator that determines whether to transfer to the next layer.

Hereinafter, a method in which the AI setting unit 238 determines the upscaling target and the AI upscaling unit 236 AI upscales the second image 135 according to the upscaling target will be described.

In an embodiment, the AI setting unit 238 may store a plurality of DNN setting information settable in the second DNN.

Here, the DNN configuration information may include information on at least one of the number of convolutional layers included in the second DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel. The plurality of DNN configuration information may each correspond to various upscale targets, and the second DNN may operate based on the DNN configuration information corresponding to a specific upscale target. The second DNN may have a different structure according to the DNN configuration information. For example, the second DNN may include three convolutional layers according to certain DNN configuration information, and the second DNN may include four convolutional layers according to other DNN configuration information.

In an embodiment, the DNN configuration information may include only parameters of a filter kernel used in the second DNN. In this case, the structure of the second DNN is not changed, but only the parameters of the internal filter kernel may be changed according to the DNN configuration information.

The AI setting unit 238 may acquire DNN setting information for AI upscaling of the second image 135 among a plurality of DNN setting information. Each of the plurality of DNN configuration information used herein is information for acquiring the third image 145 of a predetermined resolution and/or predetermined image quality, and is trained in association with the first DNN.

For example, the DNN setting information of any one of the plurality of DNN setting information is a third image 145 having a resolution twice that of the second image 135, for example, a second image of 2K (2048 * 1080). It may include information for acquiring the third image 145 of 4K (4096*2160) that is twice as large as the 2nd image 135 , and the other DNN setting information is 4 more than the resolution of the second image 135 . Information for obtaining a third image 145 of twice the resolution, for example, a third image 145 of 8K (8192 * 4320) that is 4 times larger than the second image 135 of 2K (2048 * 1080) may include

Each of the plurality of DNN setting information is created in association with the DNN setting information of the first DNN of the AI encoding device 700, and the AI setting unit 238 is set at an enlargement ratio corresponding to the reduction rate of the DNN setting information of the first DNN. Accordingly, one piece of DNN configuration information among a plurality of DNN configuration information is acquired. To this end, the AI setting unit 238 should check the information of the first DNN. In order for the AI setting unit 238 to check the information of the first DNN, the AI decoding apparatus 200 according to an embodiment receives AI data including the information of the first DNN from the AI encoding apparatus 700 .

In other words, the AI setting unit 238 uses the information received from the AI encoding device 700 to identify information targeted by the DNN setting information of the first DNN used to obtain the first image 115 , and , it is possible to obtain the DNN configuration information of the second DNN trained in association with it.

When DNN setting information for AI upscaling of the second image 135 is obtained among a plurality of DNN setting information, the DNN setting information is transmitted to the AI upscaling unit 236, and a second DNN operating according to the DNN setting information. Based on the input data may be processed.

For example, when any one DNN configuration information is obtained, the AI upscaling unit 236 includes the first convolutional layer 310 and the second convolutional layer 330 of the second DNN 300 shown in FIG. 3 . ) and the third convolutional layer 350, the number of filter kernels included in each layer and parameters of the filter kernels are set to values included in the obtained DNN configuration information.

Specifically, the AI upscaling unit 236 determines that the parameters of the filter kernel of 3 X 3 used in any one convolution layer of the second DNN shown in FIG. 4 are {1, 1, 1, 1, 1, 1 , 1, 1, 1}, and when there is a change in DNN configuration information, the parameters of the filter kernel are changed to {2, 2, 2, 2, 2, 2, 2, 2, 2, parameters included in the changed DNN configuration information. } can be replaced.

The AI setting unit 238 may acquire DNN setting information for upscaling the second image 135 among a plurality of DNN setting information based on the information included in the AI data, which is used to obtain the DNN setting information. AI data will be explained in detail.

In an embodiment, the AI setting unit 238 may obtain DNN setting information for upscaling the second image 135 among a plurality of DNN setting information, based on the difference information included in the AI data. For example, based on the difference information, the resolution of the original image 105 (eg, 4K (4096 * 2160)) is higher than the resolution of the first image 115 (eg, 2K (2048 * 1080)). When it is confirmed that the size is 2 times larger, the AI setting unit 238 may obtain DNN setting information capable of increasing the resolution of the second image 135 by 2 times.

In another embodiment, the AI setting unit 238 is based on the information related to the first image 115 included in the AI data, DNN setting information for AI upscaling the second image 135 among a plurality of DNN setting information. can be obtained. The AI setting unit 238 may determine a mapping relationship between the image related information and the DNN setting information in advance, and obtain DNN setting information mapped to the first image 115 related information.

5 is an exemplary diagram illustrating a mapping relationship between various pieces of image-related information and various pieces of DNN configuration information.

5, it can be seen that the AI encoding/AI decoding process according to an embodiment of the present disclosure does not only consider a change in resolution. As shown in FIG. 5, DNN setting information considering resolutions such as SD, HD, Full HD, bitrates such as 10Mbps, 15Mbps, and 20Mbps, and codec information such as AV1, H.264, and HEVC individually or all selection can be made. For this consideration, training in consideration of each element in the AI training process should be performed in connection with the encoding and decoding processes (see FIG. 11 ).

Accordingly, as shown in FIG. 5 according to the training content, when a plurality of DNN setting information is provided based on image-related information including a codec type and image resolution, the first image ( 115) DNN setting information for AI upscaling of the second image 135 may be acquired based on the related information.

That is, the AI setting unit 238 matches the image-related information shown on the left side of the table shown in FIG. 5 with the DNN setting information on the right side of the table, so that the DNN setting information according to the image-related information can be used.

5 , from the information related to the first image 115 , the resolution of the first image 115 is SD, and the bit rate of image data obtained as a result of the first encoding of the first image 115 is 10Mbps, , when it is confirmed that the first image 115 is first coded with the AV1 codec, the AI setting unit 238 may acquire A DNN setting information among a plurality of DNN setting information.

In addition, from the information related to the first image 115 , the resolution of the first image 115 is HD, the bit rate of image data obtained as a result of the first encoding is 15 Mbps, and the first image 115 is converted to the H.264 codec. When it is confirmed that the first encoding is performed, the AI setting unit 238 may acquire B DNN setting information among a plurality of DNN setting information.

In addition, from the information related to the first image 115 , the resolution of the first image 115 is Full HD, the bit rate of image data obtained as a result of the first encoding of the first image 115 is 20 Mbps, and the first image ( 115) is confirmed to be first encoded by the HEVC codec, the AI setting unit 238 obtains C DNN setting information among a plurality of DNN setting information, the resolution of the first image 115 is Full HD, and the first When the bit rate of the image data obtained as a result of the first encoding of the image 115 is 15 Mbps and it is confirmed that the first image 115 is first encoded with the HEVC codec, the AI setting unit 238 sets a plurality of DNN setting information Among them, DNN configuration information can be obtained. Either one of the C DNN setting information and the D DNN setting information is selected according to whether the bit rate of the image data obtained as a result of the first encoding of the first image 115 is 20 Mbps or 15 Mbps. When the first image 115 of the same resolution is first encoded with the same codec, the fact that the bit rates of the image data are different from each other means that the image quality of the reconstructed images is different from each other. Therefore, the first DNN and the second DNN may be jointly trained based on a predetermined image quality, and accordingly, the AI setting unit 238 sets the DNN according to the bit rate of the image data indicating the image quality of the second image 135 . information can be obtained.

In another embodiment, the AI setting unit 238 relates to the information (prediction mode information, motion information, quantization parameter information, etc.) provided from the first decoder 234 and the first image 115 included in the AI data. DNN configuration information for AI upscaling of the second image 135 among a plurality of pieces of DNN configuration information may be acquired in consideration of all information. For example, the AI setting unit 238 receives quantization parameter information used in the first encoding process of the first image 115 from the first decoder 234 , and receives the quantization parameter information from the AI data of the first image 115 . A bit rate of image data obtained as a result of encoding may be checked, and DNN setting information corresponding to a quantization parameter and a bit rate may be obtained. Even at the same bit rate, there may be a difference in the quality of the reconstructed image depending on the complexity of the image, and the bit rate is a value representative of the entire first image 115 to be first encoded, and the bit rate of each frame in the first image 115 is The picture quality may be different. Accordingly, when the prediction mode information, motion information, and/or quantization parameters obtainable from the first decoder 234 for each frame are considered together, the DNN setting more suitable for the second image 135 compared to using only AI data information can be obtained.

Also, depending on the implementation, the AI data may include an identifier of mutually agreed DNN configuration information. The identifier of the DNN setting information is an upscale target corresponding to the downscale target of the first DNN, and the DNN setting information trained in association between the first DNN and the second DNN so that the second image 135 can be AI upscaled. Information for distinguishing pairs of The AI setting unit 238 obtains the identifier of the DNN setting information included in the AI data, and then obtains the DNN setting information corresponding to the identifier of the DNN setting information, and the AI upscaling unit 236 receives the corresponding DNN setting information. The second image 135 may be upscaled using AI. For example, an identifier indicating each of a plurality of pieces of DNN configuration information configurable in the first DNN and an identifier indicating each of a plurality of pieces of DNN configuration information configurable in the second DNN may be predefined. In this case, the same identifier may be designated for a pair of DNN configuration information configurable to each of the first DNN and the second DNN. The AI data may include an identifier of DNN setting information set in the first DNN for AI downscaling of the original image 105 . The AI setting unit 238 that has received the AI data acquires the DNN setting information indicated by the identifier included in the AI data among a plurality of DNN setting information, and the AI upscaling unit 236 uses the corresponding DNN setting information to obtain the second The image 135 may be upscaled by AI.

Also, according to implementation, AI data may include DNN configuration information. The AI setting unit 238 may obtain DNN setting information included in AI data, and the AI upscaling unit 236 may AI upscale the second image 135 using the corresponding DNN setting information.

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 238 obtains DNN setting information by combining some selected among the lookup table values based on information included in the AI data, and the AI upscaling unit 236 uses the corresponding DNN setting information to obtain the second image (135) may be AI upscaled.

According to an embodiment, when the structure of the DNN corresponding to the upscale target is determined, the AI setting unit 238 may obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel.

As described above, the AI setting unit 238 obtains DNN setting information of the second DNN through AI data including information related to the first DNN, and the AI upscaling unit 236 sets the DNN setting information to the corresponding DNN setting information. AI upscales the second image 135 through the second DNN, which may reduce memory usage and computational amount compared to upscaling by directly analyzing the characteristics of the second image 135 .

In an embodiment, when the second image 135 consists of a plurality of frames, the AI setting unit 238 may independently obtain DNN setting information for each predetermined number of frames, or a common DNN for all frames. It is also possible to obtain setting information.

6 is a diagram illustrating a second image 135 composed of a plurality of frames.

As shown in FIG. 6 , the second image 135 may include frames corresponding to t0 to tn.

In one example, the AI setting unit 238 obtains DNN setting information of the second DNN through AI data, and the AI upscaling unit 236 AI frames corresponding to t0 to tn based on the corresponding DNN setting information. can be upscaled. That is, frames corresponding to t0 to tn may be AI upscaled based on common DNN configuration information.

In another example, the AI setting unit 238 obtains 'A' DNN setting information from AI data for some of the frames corresponding to t0 to tn, for example, frames corresponding to t0 to ta. and 'B' DNN configuration information may be obtained from AI data for frames corresponding to ta+1 to tb. Also, the AI setting unit 238 may obtain 'C' DNN setting information from AI data for frames corresponding to tb+1 to tn. In other words, the AI setting unit 238 independently obtains DNN setting information for each group including a predetermined number of frames among the plurality of frames, and the AI upscaling unit 236 independently selects the frames included in each group. AI upscaling is possible with the acquired DNN configuration information.

In another example, the AI setting unit 238 may independently obtain DNN setting information for each frame constituting the second image 135 . For example, when the second image 135 consists of three frames, the AI setting unit 238 obtains DNN setting information in relation to the first frame, and DNN setting information in relation to the second frame. and may acquire DNN configuration information in relation to the third frame. That is, DNN configuration information may be independently obtained for each of the first frame, the second frame, and the third frame. DNN setting information is obtained based on the information (prediction mode information, motion information, quantization parameter information, etc.) provided from the above-described first decoder 234 and information related to the first image 115 included in the AI data According to the method, DNN setting information may be independently obtained for each frame constituting the second image 135 . This is because mode information, quantization parameter information, and the like can be independently determined for each frame constituting the second image 135 .

In another example, the AI data may include information indicating to which frame the DNN configuration information obtained based on the AI data is valid. For example, if the AI data includes information that the DNN setting information is valid until the ta frame, the AI setting unit 238 acquires the DNN setting information based on the AI data, and the AI upscaling unit 236 is AI upscales t0 to ta frames with corresponding DNN configuration information. And, when the information that the DNN setting information is valid until the tn frame is included in other AI data, the AI setting unit 238 acquires the DNN setting information based on the other AI data, and the AI upscaling unit 236 may AI upscale ta+1 to tn frames with the obtained DNN configuration information.

Hereinafter, an AI encoding apparatus 700 for AI encoding of the original image 105 will be described with reference to FIG. 7 .

7 is a block diagram illustrating a configuration of an AI encoding apparatus 700 according to an embodiment.

Referring to FIG. 7 , the AI encoding apparatus 700 may include an AI encoding unit 710 and a transmission unit 730 . The AI encoder 710 may include an AI downscaler 712 , a first encoder 714 , a data processor 716 , and an AI setting unit 718 .

7 illustrates the AI encoder 710 and the transmitter 730 as separate devices, the AI encoder 710 and the transmitter 730 may be implemented through one processor. In this case, it may be implemented as a dedicated processor, or it may be implemented through a combination of an AP or a general-purpose processor such as a CPU or GPU and S/W. In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

Also, the AI encoder 710 and the transmitter 730 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU.

In an embodiment, the first encoding unit 714 is configured as a first processor, and the AI downscale unit 712 , the data processing unit 716 , and the AI setting unit 718 are configured as a second processor different from the first processor. In the implementation, the transmitter 730 may be implemented as a third processor different from the first processor and the second processor. The AI encoder 710 performs AI downscaling of the original image 105 and the first encoding of the first image 115 , and transmits the AI encoded data to the transmitter 730 . The transmitter 730 transmits the AI encoded data to the AI decoding apparatus 200 .

The image data includes data obtained as a result of the first encoding of the first image 115 . The image data may include data obtained based on pixel values in the first image 115 , for example, residual data that is a difference between the first image 115 and prediction data of the first image 115 . . Also, the image data includes information used in the first encoding process of the first image 115 . For example, the image data may include prediction mode information used for first encoding the first image 115 , motion information, and quantization parameter related information used for first encoding the first image 115 , etc. .

The AI data includes information enabling the AI upscaling unit 236 to AI upscale the second image 135 to an upscale target corresponding to the downscale target of the first DNN. In one example, the AI data may include difference information between the original image 105 and the first image 115 . In an example, the AI data may include information related to the first image 115 . The information related to the first image 115 is used for the resolution of the first image 115 , the bit rate of image data obtained as a result of the first encoding of the first image 115 , and the first encoding of the first image 115 . information on at least one of the specified codec types may be included.

In an embodiment, the AI data may include an identifier of mutually agreed DNN configuration information so that the second image 135 can be AI upscaled to an upscale target corresponding to the downscale target of the first DNN. .

Also, in an embodiment, the AI data may include DNN configuration information settable in the second DNN.

The AI downscaler 712 may acquire the AI downscaled first image 115 from the original image 105 through the first DNN. The AI downscaling unit 712 may AI downscale the original image 105 using the DNN setting information provided from the AI setting unit 718 . The AI setting unit 718 may determine a downscale target of the original image 105 based on a predetermined criterion.

In order to acquire the first image 115 matching the downscale target, the AI setting unit 718 may store a plurality of settable DNN setting information in the first DNN. The AI setting unit 718 obtains DNN setting information corresponding to the downscale target among a plurality of DNN setting information, and provides the obtained DNN setting information to the AI downscale unit 712 .

Each of the plurality of pieces of DNN configuration information may be trained to obtain the first image 115 having a predetermined resolution and/or a predetermined image quality. For example, the DNN setting information of any one of the plurality of DNN setting information is the first image 115 having a resolution that is 1/2 times smaller than the resolution of the original image 105, for example, 4K (4096*2160) may include information for obtaining the first image 115 of 2K (2048*1080) 1/2 times smaller than the original image 105 of A first image 115 of a resolution of 1/4 times smaller than that of a first image 115, for example, a first image 115 of 2K (2048*1080) which is 1/4 times smaller than an original image 105 of 8K (8192*4320). ) may include information for obtaining

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 718 may provide the DNN setting information obtained by combining some selected among the lookup table values to the AI downscaling unit 712 according to the downscaling target.

According to an embodiment, the AI setting unit 718 may determine the structure of the DNN corresponding to the downscale target, and obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel.

The plurality of DNN configuration information for AI downscaling of the original image 105 may have an optimized value by performing joint training of the first DNN and the second DNN. Here, each DNN configuration information includes at least one of the number of convolutional layers included in the first DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel.

The AI downscaling unit 712 sets the first DNN as the DNN setting information determined for the AI downscaling of the original image 105, and converts the first image 115 of a predetermined resolution and/or a predetermined quality to the first DNN. can be obtained through When DNN setting information for AI downscaling of the original image 105 is obtained among a plurality of DNN setting information, each layer in the first DNN may process input data based on information included in the DNN setting information. .

Hereinafter, a method for the AI setting unit 718 to determine the downscale target will be described. The downscale target may indicate, for example, how much the first image 115 with reduced resolution should be obtained from the original image 105 .

The AI setting unit 718 acquires one or more pieces of input information. In an embodiment, the input information includes a target resolution of the first image 115, a target bitrate of the image data, a bitrate type of the image data (eg, a variable bitrate type, a constant bitrate type, or an average bitrate type); Color format to which AI downscale is applied (luminance component, chrominance component, red component, green component, or blue component, etc.), codec type for first encoding of first image 115, compression history information, original image 105 at least one of a resolution of , and a type of the original image 105 .

One or more pieces of input information may include information previously stored in the AI encoding apparatus 700 or received from a user.

The AI setting unit 718 controls the operation of the AI downscale unit 712 based on the input information. In an embodiment, the AI setting unit 718 may determine a downscale target according to input information, and provide DNN setting information corresponding to the determined downscale target to the AI downscale unit 712 .

In an embodiment, the AI setting unit 718 transmits at least a portion of the input information to the first encoding unit 714 so that the first encoding unit 714 provides a bitrate of a specific value, a bitrate of a specific type, and a specific codec. may cause the first image 115 to be first encoded.

In an embodiment, the AI setting unit 718 may include a compression rate (eg, a difference in resolution between the original image 105 and the first image 115 , a target bitrate), and a compression quality (eg, a bitrate type). ), compression history information, and a downscale target may be determined based on at least one of the type of the original image 105 .

In an example, the AI setting unit 718 may determine a downscale target based on a preset or a compression rate or compression quality input from a user.

As another example, the AI setting unit 718 may determine the downscale target by using the compression history information stored in the AI encoding apparatus 700 . For example, according to the compression history information available to the AI encoding apparatus 700, the encoding quality or compression rate preferred by the user may be determined, and the downscale target may be determined according to the encoding quality determined based on the compression history information. can For example, the resolution and quality of the first image 115 may be determined according to the encoding quality that has been most frequently used according to the compression history information.

As another example, the AI setting unit 718 sets the encoding quality that has been used more than a predetermined threshold value (eg, average quality of encoding qualities that have been used more than a predetermined threshold value) according to the compression history information. A downscale target may be determined based on it.

As another example, the AI setting unit 718 may determine the downscale target based on the resolution and type (eg, file format) of the original image 105 .

In one embodiment, when the original image 105 is composed of a plurality of frames, the AI setting unit 718 independently acquires DNN setting information for each predetermined number of frames, and uses the independently acquired DNN setting information to AI downscaling. A portion 712 may be provided.

In one example, the AI setting unit 718 may divide the frames constituting the original image 105 into a predetermined number of groups, and independently obtain DNN setting information for each group. The same or different DNN configuration information may be obtained for each group. The number of frames included in the groups may be the same or different for each group.

In another example, the AI setting unit 718 may independently determine DNN setting information for each frame constituting the original image 105 . The same or different DNN configuration information may be obtained for each frame.

Hereinafter, an exemplary structure of the first DNN 800 as a basis for AI downscaling will be described.

8 is an exemplary diagram illustrating the first DNN 800 for AI downscaling of the original image 105 .

As shown in FIG. 8 , the original image 105 is input to the first convolutional layer 810 . The first convolutional layer 810 performs convolution processing on the original image 105 using 32 filter kernels having a size of 5×5. The 32 feature maps generated as a result of the convolution process are input to the first activation layer 820 . The first activation layer 820 may provide non-linear characteristics to 32 feature maps.

The first activation layer 820 determines whether to transfer sample values of the feature maps output from the first convolution layer 810 to the second convolution layer 830 . For example, some sample values among the sample values of the feature maps are activated by the first activation layer 820 and transmitted to the second convolution layer 830 , and some sample values are activated by the first activation layer 820 . It is deactivated and is not transmitted to the second convolutional layer 830 . Information indicated by the feature maps output from the first convolution layer 810 is emphasized by the first activation layer 820 .

The output 825 of the first activation layer 820 is input to the second convolution layer 830 . The second convolution layer 830 performs convolution processing on input data by using 32 filter kernels having a size of 5×5. The 32 feature maps output as a result of the convolution process are input to the second activation layer 840 , and the second activation layer 840 may provide nonlinear characteristics to the 32 feature maps.

The output 845 of the second activation layer 840 is input to the third convolution layer 850 . The third convolution layer 850 performs convolution processing on input data using one filter kernel having a size of 5×5. As a result of the convolution processing, one image may be output from the third convolutional layer 850 . The third convolutional layer 850 is a layer for outputting a final image and obtains one output by using one filter kernel. According to the example of the present disclosure, the third convolution layer 850 may output the first image 115 through the convolution operation result.

DNN setting information indicating the number of filter kernels of the first convolutional layer 810, the second convolutional layer 830, and the third convolutional layer 850 of the first DNN 800, parameters of the filter kernel, etc. There may be a plurality, and the plurality of DNN configuration information should be associated with the plurality of DNN configuration information of the second DNN. The association between the plurality of DNN configuration information of the first DNN and the plurality of DNN configuration information of the second DNN may be implemented through association learning of the first DNN and the second DNN.

8 illustrates that the first DNN 800 includes three convolutional layers 810 , 830 , 750 and two activation layers 820 , 840 , but this is only an example, and the implementation Accordingly, the number of convolutional layers and activation layers may be variously changed. Also, according to an embodiment, the first DNN 800 may be implemented through a recurrent neural network (RNN). In this case, it means changing the CNN structure of the first DNN 800 according to the example of the present disclosure to the RNN structure.

In an embodiment, the AI downscaling unit 712 may include at least one ALU for a convolution operation and an activation layer operation. The ALU may be implemented as a processor. For the convolution operation, the ALU may include a multiplier that performs a multiplication operation between the sample values of the feature map output from the original image 105 or the previous layer and the sample values of the filter kernel, and an adder that adds the result values of the multiplication. have. In addition, for the calculation of the activation layer, the ALU is a multiplier that multiplies an input sample value by a weight used in a predetermined sigmoid function, a Tanh function, or a ReLU function, and compares the multiplication result with a predetermined value to calculate the input sample value It may include a comparator that determines whether to transfer to the next layer.

Referring back to FIG. 7 , the AI setting unit 718 transmits AI data to the data processing unit 716 . The AI data includes information enabling the AI upscaling unit 236 to AI upscale the second image 135 to an upscale target corresponding to the downscale target of the first DNN. The first encoder 714 receiving the first image 115 from the AI downscaler 712 first encodes the first image 115 according to the frequency transformation-based image compression method to obtain the first image 115 . ) can reduce the amount of information it has. As a result of the first encoding through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1), image data is obtained. The image data is obtained according to a rule of a predetermined codec, that is, a syntax. For example, the image data includes residual data that is a difference between the first image 115 and the prediction data of the first image 115 , prediction mode information used to first encode the first image 115 , motion information, and Information related to a quantization parameter used to first encode the first image 115 may be included. The image data obtained as a result of the first encoding by the first encoder 714 is provided to the data processor 716 .

The data processing unit 716 generates AI encoded data including the image data received from the first encoding unit 714 and the AI data received from the AI setting unit 718 .

In an embodiment, the data processing unit 716 may generate AI-encoded data including image data and AI data in a separate state. As an example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream.

In another embodiment, the data processing unit 716 may include AI data in image data obtained as a result of the first encoding by the first encoding unit 714 and generate AI encoded data including the corresponding image data. . For example, the data processing unit 716 may generate image data in the form of one bitstream by combining a bitstream corresponding to image data and a bitstream corresponding to AI data. To this end, the data processing unit 716 may express AI data as bits having a value of 0 or 1, that is, a bitstream. In an embodiment, the data processing unit 716 may include a bitstream corresponding to AI data in supplemental enhancement information (SEI), which is an additional information area of a bitstream obtained as a result of the first encoding.

AI-encoded data is transmitted to the transmitter 730 . The transmission unit 730 transmits the AI-encoded data obtained as a result of the AI-encoding through a network. In an embodiment, the AI-encoded data is a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, and a CD-ROM and a DVD. It may be stored in a data storage medium including an optical recording medium, a magneto-optical medium such as a floptical disk, and the like.

9 is a diagram illustrating a structure of AI-encoded data 900 according to an embodiment.

As described above, the AI data 912 and the image data 932 may be separately included in the AI encoded data 900 . Here, the AI-encoded data 900 may be in a container format such as MP4, AVI, MKV, or FLV. The AI-encoded data 900 may include a metadata box 910 and a media data box 930 .

The metadata box 910 includes information about the image data 932 included in the media data box 930 . For example, the metadata box 910 may include information on the type of the first image 115 , the type of codec used to encode the first image 115 , and the playback time of the first image 115 . have. Also, the metadata box 910 may include AI data 912 . The AI data 912 may be encoded according to an encoding method provided by a predetermined container format and stored in the metadata box 910 .

The media data box 930 may include image data 932 generated according to the syntax of a predetermined image compression method.

10 is a diagram illustrating a structure of AI encoded data 1000 according to another embodiment.

Referring to FIG. 10 , AI data 1034 may be included in image data 1032 . The AI-encoded data 1000 may include a metadata box 1010 and a media data box 1030 . When the AI data 1034 is included in the image data 1032 , the metadata box 1010 contains AI Data 1034 may not be included.

The media data box 1030 includes image data 1032 including AI data 1034 . For example, the AI data 1034 may be included in the additional information area of the image data 1032 .

Hereinafter, a method of jointly training the first DNN 800 and the second DNN 300 will be described with reference to FIG. 11 .

11 is a diagram for explaining a method of training the first DNN 800 and the second DNN 300 .

In an embodiment, the AI-encoded original image 105 through the AI encoding process is restored to the third image 145 through the AI decoding process. The third image 145 and the original image 105 obtained as a result of AI decoding In order to maintain the similarity with the AI encoding process and the AI decoding process, correlation is required. That is, information lost in the AI encoding process should be able to be restored in the AI decoding process, and for this purpose, joint training between the first DNN 800 and the second DNN 300 is required.

For accurate AI decoding, it is ultimately necessary to reduce the quality loss information 1130 corresponding to the comparison result between the third training image 1104 and the original training image 1101 shown in FIG. 11 . Accordingly, the quality loss information 1130 is used for both training of the first DNN 800 and the second DNN 300 .

First, the training process shown in FIG. 11 will be described.

In FIG. 11 , an original training image 1101 is an image subject to AI downscale, and a first training image 1102 is an AI downscaled image from the original training image 1101 . it's a video Also, the third training image 1104 is an AI upscaled image from the first training image 1102 .

The original training image 1101 includes a still image or a moving image composed of a plurality of frames. In an embodiment, the original training image 1101 may include a still image or a luminance image extracted from a moving picture including a plurality of frames. Also, according to an embodiment, the original training image 1101 may include a still image or a patch image extracted from a moving image including a plurality of frames. When the original training image 1101 consists of a plurality of frames, the first training image 1102 , the second training image and the third training image 1104 also include a plurality of frames. When a plurality of frames of the original training image 1101 are sequentially input to the first DNN 800 , the first training image 1102 and the second training image are performed through the first DNN 800 and the second DNN 300 . and a plurality of frames of the third training image 1104 may be sequentially acquired.

For joint training of the first DNN 800 and the second DNN 300 , an original training image 1101 is input to the first DNN 800 . The original training image 1101 input to the first DNN 800 is AI downscaled and output as the first training image 1102 , and the first training image 1102 is input to the second DNN 300 . A third training image 1104 is output as a result of AI upscaling for the first training image 1102 .

Referring to FIG. 11 , a first training image 1102 is being input to the second DNN 300 . According to an embodiment, the first training image 1102 is obtained through a first encoding and a first decoding process. A second training image may be input to the second DNN 300 . In order to input the second training image to the second DNN, any one of MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1 codecs may be used. Specifically, in the first encoding of the first training image 1102 and the first decoding of image data corresponding to the first training image 1102, MPEG-2, H.264, MPEG-4, HEVC, VC-1 , VP8, VP9, and any one codec of AV1 may be used.

Referring to FIG. 11 , the legacy downscaled reduced training image 1103 is obtained from the original training image 1101 separately from the output of the first training image 1102 through the first DNN 800 . Here, the legacy downscale may include at least one of a bilinear scale, a bicubic scale, a lanczos scale, and a stair step scale.

In order to prevent the structural features of the first image 115 from greatly deviate based on the structural features of the original image 105, a reduced training image 1103 that preserves the structural features of the original training image 1101 is acquired. .

Before the training proceeds, the first DNN 800 and the second DNN 300 may be set with predetermined DNN configuration information. As training progresses, structural loss information 1110 , complexity loss information 1120 , and quality loss information 1130 may be determined.

The structural loss information 1110 may be determined based on a comparison result between the reduced training image 1103 and the first training image 1102 . In one example, the structural loss information 1110 may correspond to a difference between the structural information of the reduced training image 1103 and the structural information of the first training image 1102 . The structural information may include various features that can be extracted from the image, such as luminance, contrast, and histogram of the image. The structural loss information 1110 indicates to what extent the structural information of the original training image 1101 is maintained in the first training image 1102 . As the structural loss information 1110 is smaller, the structural information of the first training image 1102 is similar to the structural information of the original training image 1101 .

The complexity loss information 1120 may be determined based on the spatial complexity of the first training image 1102 . In an example, as the spatial complexity, a total variance value of the first training image 1102 may be used. The complexity loss information 1120 is related to a bit rate of image data obtained by first encoding the first training image 1102 . The smaller the complexity loss information 1120 is, the smaller the bit rate of image data is defined.

The quality loss information 1130 may be determined based on a comparison result between the original training image 1101 and the third training image 1104 . The quality loss information 1130 includes an L1-norm value, an L2-norm value, a Structural Similarity (SSIM) value, and a Peak Signal-To (PSNR-HVS) value for the difference between the original training image 1101 and the third training image 1104 . It may include at least one of a Noise Ratio-Human Vision System) value, a Multiscale SSIM (MS-SSIM) value, a Variance Inflation Factor (VIF) value, and a Video Multimethod Assessment Fusion (VMAF) value. The quality loss information 1130 indicates how similar the third training image 1104 is to the original training image 1101 . As the quality loss information 1130 is smaller, the third training image 1104 is more similar to the original training image 1101 .

11, structural loss information 1110, complexity loss information 1120, and quality loss information 1130 are used for training of the first DNN 800, and the quality loss information 1130 is the second DNN ( 300) is used for training. That is, the quality loss information 1130 is used for both training of the first DNN 800 and the second DNN 300 .

The first DNN 800 may update parameters such that the final loss information determined based on the structural loss information 1110 , the complexity loss information 1120 , and the quality loss information 1130 is reduced or minimized. Also, the second DNN 300 may update the parameter so that the quality loss information 1130 is reduced or minimized.

Final loss information for training of the first DNN 800 and the second DNN 300 may be determined as in Equation 1 below.

[Equation 1]

In Equation 1, LossDS represents final loss information to be reduced or minimized for training of the first DNN 800, and LossUS represents final loss information to be reduced or minimized for training of the second DNN 300. indicates. Also, a, b, c, and d may correspond to predetermined weights.

That is, the first DNN 800 updates parameters in a direction in which LossDS of Equation 1 is decreased, and the second DNN 300 updates parameters in a direction in which LossUS is decreased. When the parameters of the first DNN 800 are updated according to the LossDS derived in the training process, the first training image 1102 obtained based on the updated parameters is different from the first training image 1102 in the previous training process. and, accordingly, the third training image 1104 is also different from the third training image 1104 in the previous training process. When the third training image 1104 is different from the third training image 1104 in the previous training process, the quality loss information 1130 is also newly determined, and accordingly, the second DNN 300 updates the parameters. When the quality loss information 1130 is newly determined, since LossDS is also newly determined, the first DNN 800 updates parameters according to the newly determined LossDS. That is, the parameter update of the first DNN 800 causes the parameter update of the second DNN 300 , and the parameter update of the second DNN 300 causes the parameter update of the first DNN 800 . In other words, since the first DNN 800 and the second DNN 300 are jointly trained through sharing of the quality loss information 1130 , the parameters of the first DNN 800 and the parameters of the second DNN 300 are They can be optimized in relation to each other.

Referring to Equation 1, it can be seen that LossUS is determined according to quality loss information 1130, but this is an example, and LossUS is at least one of structural loss information 1110 and complexity loss information 1120, It may be determined based on the quality loss information 1130 .

Previously, it has been described that the AI setting unit 238 of the AI decoding apparatus 200 and the AI setting unit 718 of the AI encoding apparatus 700 store a plurality of DNN setting information, the AI setting unit 238 and the AI A method of training each of the plurality of DNN configuration information stored in the setting unit 718 will be described.

As described in relation to Equation 1, in the case of the first DNN 800, the degree of similarity between the structural information of the first training image 1102 and the structural information of the original training image 1101 (structural loss information 1110) ), the bit rate (complexity loss information 1120) of the image data obtained as a result of the first encoding of the first training image 1102 and the difference between the third training image 1104 and the original training image 1101 (quality loss) The parameter is updated in consideration of the information 1130).

In detail, the first training image 1102 similar to the structural information of the original training image 1101 and having a small bit rate of image data obtained when the first encoding is performed can be obtained, and at the same time, the first training image The parameters of the first DNN 800 may be updated so that the second DNN 300 that AI upscales 1102 may acquire a third training image 1104 similar to the original training image 1101 .

By adjusting the weights of a, b, and c in Equation 1, the directions in which the parameters of the first DNN 800 are optimized are different. For example, when the weight of b is determined to be high, the parameter of the first DNN 800 may be updated with more importance on lowering the bit rate than the quality of the third training image 1104 . In addition, when the weight of c is determined to be high, it is more important to increase the quality of the third training image 1104 than to increase the bit rate or to maintain the structural information of the original training image 1101. 1 A parameter of the DNN 800 may be updated.

Also, the direction in which parameters of the first DNN 800 are optimized may be different according to the type of a codec used to first encode the first training image 1102 . This is because, depending on the type of codec, the second training image to be input to the second DNN 300 may vary.

That is, the parameters of the first DNN 800 and the parameters of the second DNN 300 are linked based on the weight a, the weight b, the weight c, and the type of the codec for the first encoding of the first training image 1102 . so it can be updated. Therefore, when each of the weight a, the weight b, and the weight c is determined as a predetermined value and the type of the codec is determined as the predetermined type, the first DNN 800 and the second DNN 300 are trained, they are linked to each other. Optimized parameters of the first DNN 800 and parameters of the second DNN 300 may be determined.

Then, when the first DNN 800 and the second DNN 300 are trained after changing the weight a, the weight b, the weight c, and the type of the codec, the parameters of the first DNN 800 are optimized in connection with each other. and parameters of the second DNN 300 may be determined. In other words, when the first DNN 800 and the second DNN 300 are trained while changing the respective values of the weight a, the weight b, the weight c, and the type of the codec, the plurality of DNN setting information trained in connection with each other is transmitted to the first It may be determined from the DNN 800 and the second DNN 300 .

As described above with reference to FIG. 5 , the plurality of DNN configuration information of the first DNN 800 and the second DNN 300 may be mapped to the first image related information. To establish such a mapping relationship, a first training image 1102 output from the first DNN 800 is first encoded with a specific codec according to a specific bit rate, and a bitstream obtained as a result of the first encoding is first decoded. The obtained second training image may be input to the second DNN 300 . That is, after setting the environment so that the first training image 1102 of a specific resolution is first encoded at a specific bit rate by a specific codec, by training the first DNN 800 and the second DNN 300 , the first DNN setting mapped to the resolution of the training image 1102, the type of codec used for the first encoding of the first training image 1102, and the bitrate of the bitstream obtained as a result of the first encoding of the first training image 1102 The information pair can be determined. The resolution of the first training image 1102, the type of codec used for the first encoding of the first training image 1102, and the bitrate of the bitstream obtained according to the first encoding of the first training image 1102 are varied. , a mapping relationship between a plurality of DNN configuration information of the first DNN 800 and the second DNN 300 and the first image related information may be determined.

12 is a diagram for explaining a training process of the first DNN 800 and the second DNN 300 by the training device 1200 .

The training of the first DNN 800 and the second DNN 300 described with reference to FIG. 11 may be performed by the training apparatus 1200 . The training device 1200 includes a first DNN 800 and a second DNN 300 . The training device 1200 may be, for example, the AI encoding device 700 or a separate server. The DNN setting information of the second DNN 300 obtained as a result of training is stored in the AI decoding apparatus 200 .

Referring to FIG. 12 , the training apparatus 1200 initially sets DNN configuration information of the first DNN 800 and the second DNN 300 ( S1240 , S1245 ). Accordingly, the first DNN 800 and the second DNN 300 may operate according to predetermined DNN configuration information. The DNN setting information includes at least the number of convolution layers included in the first DNN 800 and the second DNN 300, the number of filter kernels for each convolutional layer, the size of filter kernels for each convolutional layer, and parameters of each filter kernel. It can contain information about one.

The training apparatus 1200 inputs the original training image 1101 into the first DNN 800 (S1250). The original training image 1101 may include at least one frame constituting a still image or a moving image.

The first DNN 800 processes the original training image 1101 according to the initially set DNN setting information, and outputs the AI downscaled first training image 1102 from the original training image 1101 ( S1255 ). 12 shows that the first training image 1102 output from the first DNN 800 is directly input to the second DNN 300 , but the first training image 1102 output from the first DNN 800 . ) may be input to the second DNN 300 by the training device 1200 . Also, the training apparatus 1200 may first encode and first decode the first training image 1102 using a predetermined codec, and then input the second training image to the second DNN 300 .

The second DNN 300 processes the first training image 1102 or the second training image according to the initially set DNN setting information, and the AI upscaled third from the first training image 1102 or the second training image. A training image 1104 is output (S1260).

The training apparatus 1200 calculates complexity loss information 1120 based on the first training image 1102 (S1265).

The training apparatus 1200 calculates structural loss information 1110 by comparing the reduced training image 1103 with the first training image 1102 ( S1270 ).

The training apparatus 1200 calculates quality loss information 1130 by comparing the original training image 1101 and the third training image 1104 (S1275).

The first DNN 800 updates the initially set DNN configuration information through a back propagation process based on the final loss information (S1280). The training apparatus 1200 may calculate final loss information for training the first DNN 800 based on the complexity loss information 1120 , the structural loss information 1110 , and the quality loss information 1130 .

The second DNN 300 updates the initially set DNN configuration information through a reverse transcription process based on quality loss information or final loss information (S1285). The training apparatus 1200 may calculate final loss information for training of the second DNN 300 based on the quality loss information 1130 .

Thereafter, the training apparatus 1200, the first DNN 800, and the second DNN 300 update the DNN configuration information while repeating processes S250 to S1285 until the final loss information is minimized. In this case, during each iteration process, the first DNN 800 and the second DNN operate according to the DNN configuration information updated in the previous process.

Table 1 below shows the effects of AI encoding and AI decoding of the original image 105 and encoding and decoding of the original image 105 using HEVC according to an embodiment of the present disclosure.

[Table 1]

As can be seen from Table 1, the subjective picture quality when AI encoding and AI decoding of contents consisting of 300 frames of 8K resolution according to an embodiment of the present disclosure is higher than the subjective picture quality when encoding and decoding using HEVC. , it can be seen that the bit rate is reduced by more than 50%.

In the following description, 'original image' refers to an image to be subjected to AI encoding, and 'first image' refers to AI downscale or AI one-to-one of the original image according to the resolution determination result of the original image in the AI encoding process. It means an image obtained as a result of preprocessing. In addition, the 'second image' refers to an image obtained through the first decoding in the AI decoding process, and the 'third image' refers to an image obtained by AI upscaling the second image in the AI decoding process.

In addition, in the following description, 'AI downscaling' refers to processing of reducing the resolution of an image when the input image is high-resolution based on AI, and 'AI 1-to-1 pre-processing' is when the input image is low-resolution based on AI. For AI upscaling of the input image, it refers to the process of preserving or enhancing the detailed features of the input image while maintaining the image resolution at one-to-one level, and the 'first encoding' is encoding by a frequency transformation-based image compression method. means processing. In addition, 'first decoding' refers to a decoding process by a frequency transformation-based image restoration method, and 'AI upscaling' refers to a process of increasing the resolution of an image based on AI.

13 is a diagram for explaining an artificial intelligence (AI) encoding process and an AI decoding process according to an embodiment.

As shown in FIG. 13 , according to an embodiment of the present disclosure, the size of the resolution of the original image 1305 is determined ( 1300 ).

When the size of the original image 1305 is greater than a predetermined value, the downscaled first image 1345 is obtained by AI downscaling 1320 of the original image 1335 having a high resolution. In addition, since the first encoding 1330 and the first decoding 1340 are performed on the first image 1345 of a relatively small resolution, the first encoding ( 1330) and the first decoding 1340 may be significantly reduced compared to the case of performing the processing bit rate.

When the size of the original image 1305 is less than or equal to a predetermined value, AI 1-to-1 pre-processing 1310 on the low-resolution original image 1315 to obtain the 1-to-1 pre-processed first image 1325 do. In addition, since the first encoding 1330 and the first decoding 1340 are performed on the first image 1325 in which the specific features of the low-resolution original image 1315 are emphasized, the first encoding 1330 and the first decoding 1340 are performed. It can compensate for information that is difficult to restore. An image with a low resolution has a large amount of information relative to the same area (for example, a 32x32 patch unit) compared to an image with a high resolution, so the complexity is large. It may be lost, resulting in image quality degradation. Therefore, in the case of a low-resolution original image, it is necessary to preserve detailed features of the original image through one-to-one pre-processing to increase the memorization at the time of upscaling.

Referring to FIG. 13 , according to an exemplary embodiment, in the AI encoding process, the size of the resolution of the original image 1305 is first determined ( 1300 ).

If the size of the original image 1305 is larger than a predetermined value, AI downscales the original image 1335 with a high resolution 1320 to obtain a downscaled first image 1345, and the first image 1345 is A first encoding (1330) is performed. In the AI decoding process, AI encoding data including AI data and image data obtained as a result of AI encoding is received, a second image 1355 is obtained through a first decoding 1340, and the second image 1355 is A third image 1365 is acquired by AI upscaling 1350 .

On the other hand, if the size of the original image 1305 is less than or equal to a predetermined value, AI 1-to-1 pre-processing 1310 of the low-resolution original image 1310 to obtain the 1-to-1 pre-processed first image 1325, The first image is first encoded (1330). In the AI decoding process, AI encoding data including AI data and image data obtained as a result of AI encoding is received, a second image 1355 is obtained through a first decoding 1340, and the second image 1355 is A third image 1365 is acquired by AI upscaling 1350 .

Looking at the AI encoding process in more detail, when an original image is input, the resolution of the original image is first determined. When the resolution of the original image is greater than a predetermined value, the AI downscales the original image to obtain a first image of a predetermined resolution or a predetermined quality. At this time, the AI downscaling 1320 is performed based on AI, the AI for the AI downscaling 1320 must be trained in connection with the AI for the AI upscaling 1350 of the second image 1355 (joint trained) do. Because, when the AI for the AI downscaling 1320 and the AI for the AI upscaling 1350 are separately trained, between the original image 1335 that is the AI encoding target and the third image 1365 restored through AI decoding because the difference between When the resolution of the original image 1305 is less than or equal to a predetermined value, the original image 1315 is pre-processed by AI one-to-one to obtain a first image 1325 in which the detailed features of the small-resolution original image 1315 are preserved. (1310) do. At this time, the AI one-to-one pre-processing is performed based on AI, and the AI for the AI one-to-one pre-processing must be jointly trained in connection with the AI for the AI upscaling (1350). Because, when AI for AI one-to-one preprocessing 1310 and AI for AI upscaling 1350 are trained separately, the third image 1365 upscaled through AI decoding is the original image ( 1315) may not be included. In this case, the parameters of the AI for the AI upscaling 1350 are used as they are by training in conjunction with the AI for the AI downscaling 1320 . That is, in the joint training of AI for AI one-to-one pre-processing and AI for AI upscaling (1350), the parameters of AI for AI upscaling (1350) are not updated, and AI for AI downscaling and It is fixed with parameters obtained through linkage training. Specifically, AI for AI upscaling (1350) and AI for AI downscaling (1320) are firstly trained in conjunction, and one AI using the parameters of AI for AI upscaling (1350) obtained as a result of training as it is. The AI for the 1 pre-processing 1310 and the AI for the AI upscaling 1350 are jointly trained, and the parameters of the AI for the AI 1-to-1 pre-processing 1310 are obtained. The parameters of AI for AI upscaling (1350) are trained in conjunction with AI for AI downscaling (1320) and by using the obtained parameters to be fixed, while maintaining the performance of AI for AI upscaling (1350), up When scaling, parameters of AI for AI one-to-one preprocessing 1310 that can reflect detailed features of low-resolution images may be obtained. When the original image is low-resolution, by using the parameters of AI for AI upscaling that have been trained in conjunction with AI for AI downscaling as it is, and training in conjunction with AI for AI one-to-one preprocessing, for AI one-to-one preprocessing AI parameters have the effect of simulating AI parameters for AI downscaling. Therefore, by providing AI parameters for AI one-to-one preprocessing that simulates AI parameters for AI downscaling by the content provider for the low-resolution original video, the content provider does not need to make the low-resolution original video into a high-resolution video. , it is possible to effectively upscale a low-resolution original image to a high-resolution image without degrading the image quality.

In an embodiment of the present disclosure, AI data may be used in order to maintain this linkage in the AI encoding process and the AI decoding process. Therefore, the AI data obtained through the AI encoding process should include information indicating the upscale target, and in the AI decoding process, the second image 1355 is AI upscaled ( 1350) should be done.

AI for AI one-to-one preprocessing 1310, AI for AI downscaling 1320, and AI for AI upscaling 1350 may be implemented as a deep neural network (DNN). As will be described later with reference to FIG. 16 , since the first DNN and the second DNN are jointly trained through sharing of loss information under a predetermined target, when the resolution of the original image is greater than a predetermined value, the AI encoding apparatus uses the first DNN and 2 DNN provides target information used when joint training is performed to the AI decoding device, and the AI decoding device upscales the AI to image quality and/or resolution targeting the second image 1355 based on the received target information. 1350) can be done.

The third DNN is trained in association with the second DNN after the first DNN and the second DNN are first trained, and then the DNN configuration information of the second DNN is fixed. Since the third DNN and the second DNN are jointly trained under a predetermined target using the second DNN of the fixed DNN setting information after the joint training of the first DNN and the second DNN, when the resolution of the original image is less than a predetermined value, AI The encoding device provides target information used when the third DNN and the second DNN are jointly trained to the AI decoding device, and the AI decoding device targets the second image 1335 based on the received target information. Alternatively, AI may be upscaled 1350 to resolution.

When the first encoding 1330 and the first decoding 1340 shown in FIG. 13 are described in detail, the resolution of the original image is determined. In the first image 1345 that is AI downscaled 1320 from the image 1335, the amount of information is reduced through the first encoding 1330, and when the resolution of the original image 1305 is less than or equal to a predetermined value, the original image with a small resolution The amount of information of the first image 1325 that has been subjected to AI one-to-one pre-processing 1310 from the image 1315 may be reduced through the first encoding 1330 . The first encoding 1330 includes a process of generating prediction data by predicting the first images 1325 and 1345, a process of generating residual data corresponding to a difference between the first images 1325 and 1345 and the prediction data; It may include a process of transforming the residual data, which is a spatial domain component, into a frequency domain component, a process of quantizing the residual data transformed into a frequency domain component, and a process of entropy encoding the quantized residual data. Such a first encoding process 1230 is MPEG-2, H.264 Advanced Video Coding (AVC), MPEG-4, High Efficiency Video Coding (HEVC), VC-1, VP8, VP9, and AOMedia Video 1 (AV1). It can be implemented through one of the image compression methods using frequency transformation, etc.

The second image 1355 corresponding to the first images 1325 and 1345 may be reconstructed through the first decoding 1340 of the image data. The first decoding 1340 includes a process of generating quantized residual data by entropy-decoding image data, a process of inverse quantizing the quantized residual data, a process of transforming the residual data of a frequency domain component into a spatial domain component, a process of predicting data It may include a process of generating , and a process of reconstructing the second image 1355 using prediction data and residual data. The first decoding 1340 process as described above is an image compression using frequency conversion such as MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1 used in the first encoding 1330 process. It may be implemented through an image restoration method corresponding to one of the methods.

The AI-encoded data obtained through the AI encoding process is related to the image data obtained as a result of the first encoding 1330 of the first images 1325 and 1345 and the AI downscale 1320 of the original image 1335 having a high resolution. AI data or AI data related to the AI one-to-one preprocessing 1310 of the original image 1315 having a small resolution may be included. The image data may be used in the first decoding 1340 process, and the AI data may be used in the AI upscaling 1350 process.

The image data may be transmitted in the form of a bitstream. The image data is data obtained based on pixel values in the first images 1325 and 1345 , for example, a residual that is a difference between the first images 1325 and 1345 and prediction data of the first images 1325 and 1345 . It may contain data. Also, the image data includes information used in the first encoding 1330 of the first images 1325 and 1345 . For example, the image data includes prediction mode information used for the first encoding 1330 of the first images 1325 and 1345 , motion information, and quantization parameter related information used in the first encoding 1230 , etc. may include. The image data is the image compression method used in the first encoding (1230) process among the image compression methods using frequency conversion, such as MPEG-2, H.264 AVC, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1. It may be generated according to a rule, for example, a syntax.

AI data is used for AI upscaling 1350 based on the second DNN. As described above, after the first DNN and the second DNN are jointly trained, the third DNN and the second DNN are further jointly trained using the second DNN of the obtained DNN setting information, so the AI data is the second It includes information enabling accurate AI upscaling 1350 of the second image 1355 via DNN to be performed. In the AI decoding process, AI upscaling 1365 may be performed to a resolution and/or image quality targeting the second image 1355 based on AI data.

AI data may be transmitted together with image data in the form of a bitstream. Alternatively, according to an embodiment, AI data may be transmitted separately from image data in the form of frames or packets.

Alternatively, according to an embodiment, AI data may be transmitted while being included in image data.

The image data and AI data may be transmitted through the same network or different networks.

The AI decoding apparatus for AI decoding according to FIG. 13 may have the same configuration as the AI decoding apparatus of FIG. 2 described above.

Referring to FIG. 2 , the AI decoding apparatus 200 according to an embodiment includes a receiving unit 210 and an AI decoding unit 230 . The AI decoding unit 230 may include a parsing unit 232 , a first decoding unit 234 , an AI upscaling unit 236 , and an AI setting unit 238 .

Although the receiver 210 and the AI decoder 230 are illustrated as separate devices in FIG. 2 , the receiver 210 and the AI decoder 230 may be implemented through one processor. In this case, the receiving unit 210 and the AI decoding unit 230 may be implemented as a dedicated processor, and a general-purpose processor such as an application processor (AP), a central processing unit (CPU), or a graphic processing unit (GPU) and S/W It may be implemented through a combination of In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

The receiver 210 and the AI decoder 230 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU. In one embodiment, the receiving unit 210 is implemented as a first processor, the first decoding unit 234 is implemented as a second processor different from the first processor, the parsing unit 232, the AI upscaling unit 236 and the AI setting unit 238 may be implemented as a third processor different from the first processor and the second processor.

The receiving unit 210 receives AI encoded data obtained as a result of AI encoding. As an example, the AI-encoded data may be a video file having a file format such as mp4 or mov.

The receiver 210 may receive AI-encoded data transmitted through a network. The receiving unit 210 outputs the AI encoded data to the AI decoding unit 230 .

In one embodiment, the AI-encoded data is a hard disk, magnetic media such as floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks. It may be obtained from a data storage medium including an optical medium).

The parsing unit 232 parses the AI encoded data and transmits image data generated as a result of the first encoding of the first images 1325 and 1345 to the first decoder 234 , and transmits the AI data to the AI setting unit 238 . ) to pass

In an embodiment, the parsing unit 232 may parse the image data and the AI data separately included in the AI encoded data. The parsing unit 232 may read the header in the AI-encoded data to distinguish the AI data and the image data included in the AI-encoded data. In one example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream.

The structure of AI-encoded data including separated AI data and image data is omitted as described above with reference to FIG. 9 .

In another embodiment, the parsing unit 232 parses the image data from the AI encoded data, extracts the AI data from the image data, transmits the AI data to the AI setting unit 238, and first decodes the remaining image data. may be transmitted to the unit 234 . That is, AI data may be included in image data. For example, AI data may be included in supplemental enhancement information (SEI), which is an additional information area of a bitstream corresponding to image data. The structure of the AI-encoded data including the image data including the AI data has been described above with reference to FIG. 10, and thus will be omitted.

In another embodiment, the parsing unit 232 divides the bitstream corresponding to the image data into a bitstream to be processed by the first decoding unit 234 and a bitstream corresponding to AI data, and divides each divided bitstream It may output to the first decoding unit 234 and the AI setting unit 238 .

The parsing unit 232 obtains image data included in the AI encoded data through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1). It can also be confirmed that it is the image data that has been processed. In this case, the corresponding information may be transmitted to the first decoder 234 so that the image data can be processed by the identified codec.

The first decoder 234 reconstructs the second image 1355 corresponding to the first images 1325 and 1345 based on the image data received from the parser 232 . The second image 1355 obtained by the first decoder 234 is provided to the AI upscaler 236 .

According to an embodiment, first decoding-related information such as prediction mode information, motion information, and quantization parameter information may be provided from the first decoding unit 234 to the AI setting unit 238 . The first decoding related information may be used to obtain DNN configuration information.

The AI data provided to the AI setting unit 238 of the AI decoding apparatus for AI decoding according to FIG. 13 includes information enabling AI upscaling of the second image 1355 . In this case, the upscale target should correspond to the downscale of the first DNN or the one-to-one preprocessing of the third DNN. Accordingly, the AI data should include information that can confirm the downscale target of the first DNN or information that can identify the one-to-one preprocessing target of the third DNN.

Specifically exemplifying the information included in the AI data, information on the difference between the resolution of the high-resolution original image 1335 and the resolution of the downscaled first image 1345, the downscaled first image 1345 related information, There is resolution information of the first image 1325 preprocessed one-to-one from the original image 1315 having a small resolution, and information related to the first image 1325 that has been pre-processed one-to-one.

The difference information may be expressed as information about a resolution conversion degree of the downscaled first image 1345 compared to the original image 1335 having a high resolution (eg, resolution conversion rate information). Also, since the resolution of the first image 1345 is known through the resolution of the reconstructed second image 1355 and the degree of resolution conversion can be confirmed through this, the difference information can be expressed only with the resolution information of the original image 1335 . may be Here, the resolution information may be expressed as a horizontal/vertical screen size, or it may be expressed as a ratio (16:9, 4:3, etc.) and a size of one axis. In addition, when there is preset resolution information, it may be expressed in the form of an index or a flag.

In addition, the downscaled first image 1345 related information may include a bit rate of image data obtained as a result of the first encoding of the first image 1345 and a codec type used during the first encoding of the first image 1345 . It may include information about at least one.

Also, since the resolution information of the first image 1325 preprocessed one-to-one is the same as the original image 1315 whose resolution is determined to be smaller than a predetermined value by the original image resolution determination 1300, the resolution information of the first image 1325 having a smaller resolution An upscale target may be determined according to a predetermined criterion according to the resolution of the image 1325 . For example, if the resolution size of the first image pre-processed one-to-one is 2K, the upscale target may be determined to be 4K. Also, the resolution information of the first image 1325 that has been pre-processed one-to-one may be resolution difference information. Specifically, since the one-to-one preprocessed first image 1325 and the original image 1315 having a small resolution have the same resolution, the resolution difference information may indicate 0. In addition, since the resolution of the first image 1325 is known through the resolution of the reconstructed second image 1355 and the degree of resolution conversion can be confirmed through this, the resolution information may be expressed only with the resolution information of the original image 1315 . may

And, similarly, the information related to the first image 1325 that has been pre-processed one-to-one is used for the bit rate of the image data obtained as a result of the first encoding of the first image 1325 and the first encoding of the first image 1325 . information on at least one of the specified codec types may be included.

The AI setting unit 238 may include at least one of the difference information included in the AI data and the downscaled first image 1345 related information, or resolution information (or resolution difference of 0) of the first image 1325 included in the AI data. The upscaling target of the second image 1355 may be determined based on at least one of (difference information indicating that . The upscaling target may indicate, for example, to what resolution the second image 1355 should be upscaled. When the upscaling target is determined, the AI setting unit 238 AI upscales the second image 1355 through the second DNN to obtain a third image 1365 corresponding to the upscaling target.

The AI upscaling process through the second DNN is omitted as described above with reference to FIGS. 3 and 4 .

Hereinafter, a method in which the AI setting unit 238 determines the upscaling target and the AI upscaling unit 236 AI upscales the second image 1355 according to the upscaling target will be described.

In an embodiment, the AI setting unit 238 may store a plurality of DNN setting information settable in the second DNN.

Here, the DNN configuration information may include information on at least one of the number of convolutional layers included in the second DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel. The plurality of DNN configuration information may each correspond to various upscale targets, and the second DNN may operate based on the DNN configuration information corresponding to a specific upscale target. The second DNN may have a different structure according to the DNN configuration information. For example, the second DNN may include three convolutional layers according to certain DNN configuration information, and the second DNN may include four convolutional layers according to other DNN configuration information.

In an embodiment, the DNN configuration information may include only parameters of a filter kernel used in the second DNN. In this case, the structure of the second DNN is not changed, but only the parameters of the internal filter kernel may be changed according to the DNN configuration information.

The AI setting unit 238 may acquire DNN setting information for AI upscaling of the second image 1355 among a plurality of DNN setting information. Each of the plurality of DNN setting information used here is information for obtaining the third image 1365 of a predetermined resolution and/or predetermined quality, and is trained in association with the first DNN and then trained in association with the third DNN. .

For example, the DNN setting information of any one of the plurality of DNN setting information is a third image 1365 having a resolution twice that of the second image 1355, for example, a second image of 2K (2048 * 1080). It may include information for acquiring the third image 1365 of 4K (4096*2160) that is twice as large as the 2nd image 1355, and the other DNN setting information is 4 more than the resolution of the second image 1255. Information for obtaining a third image 1365 with a resolution that is twice as large, for example, a third image 1365 of 8K (8192 * 4320) that is four times larger than the second image 1355 of 2K (2048 * 1080) may include

Each of the plurality of DNN configuration information is created in association with the DNN configuration information of the first DNN of the AI encoding apparatus 1500, and then is associated with the DNN configuration information of the third DNN. The AI setting unit 238 obtains one DNN setting information from among a plurality of DNN setting information according to an enlargement ratio corresponding to a reduction rate of the DNN setting information of the first DNN. To this end, the AI setting unit 238 should check the information of the first DNN. In order for the AI setting unit 238 to check the information of the first DNN, the AI decoding apparatus 200 according to an embodiment receives AI data including the information of the first DNN from the AI encoding apparatus 1500 .

Also, the AI setting unit 238 obtains one DNN setting information among a plurality of DNN setting information according to the resolution information of the first image 1325 of the DNN setting information of the third DNN. To this end, the AI setting unit 238 should check the information of the third DNN. In order for the AI setting unit 238 to check the information of the third DNN, the AI decoding apparatus 200 according to an embodiment receives AI data including the information of the third DNN from the AI encoding apparatus 1500 .

In other words, the AI setting unit 238 uses the information received from the AI encoding apparatus 1500 to determine the DNN setting information of the first DNN or the third DNN used to acquire the first images 1325 and 1345 . It is possible to confirm the target information and obtain the DNN setting information of the second DNN trained in association with it.

When DNN setting information for AI upscaling of the second image 1355 is obtained among a plurality of DNN setting information, the DNN setting information is transmitted to the AI upscaling unit 236, and a second DNN operating according to the DNN setting information. Based on the input data may be processed.

For example, when any one DNN configuration information is obtained, the AI upscaling unit 236 includes the first convolutional layer 310 and the second convolutional layer 330 of the second DNN 300 shown in FIG. 3 . ) and the third convolutional layer 350, the number of filter kernels included in each layer and parameters of the filter kernels are set to values included in the obtained DNN configuration information.

Specifically, the AI upscaling unit 236 determines that the parameters of the filter kernel of 3 X 3 used in any one convolution layer of the second DNN shown in FIG. 4 are {1, 1, 1, 1, 1, 1 , 1, 1, 1}, and when there is a change in DNN configuration information, the parameters of the filter kernel are changed to {2, 2, 2, 2, 2, 2, 2, 2, 2, parameters included in the changed DNN configuration information. } can be replaced.

The AI setting unit 238 may acquire DNN setting information for upscaling the second image 1355 among a plurality of DNN setting information based on information included in the AI data, which is used to obtain the DNN setting information. AI data will be explained in detail.

In an embodiment, the AI setting unit 238 may obtain DNN setting information for upscaling the second image 1355 from among a plurality of DNN setting information, based on the difference information included in the AI data. For example, based on the difference information, the resolution of the original image 1335 (eg, 4K (4096 * 2160)) is higher than the resolution of the first image 1345 (eg, 2K (2048 * 1080)). When it is confirmed that it is twice as large, the AI setting unit 238 may obtain DNN setting information capable of increasing the resolution of the second image 1355 by two times.

In another embodiment, the AI setting unit 238 upscales the second image 1355 among the plurality of DNN setting information based on the resolution information of the one-to-one preprocessed first image 1325 included in the AI data. It is possible to obtain DNN configuration information for For example, since the resolution of the one-to-one preprocessed first image 1325 is the same as that of the original image 1215 determined to be smaller than a predetermined value, according to the resolution information of the first image 1225 , a predetermined standard An upscale target may be determined by For example, if the resolution size of the first image pre-processed one-to-one is 2K, the upscale target may be determined to be 4K.

In another embodiment, the AI setting unit 238 is a DNN for AI upscaling the second image 1355 among a plurality of DNN setting information based on the first image 1325 and 1345 related information included in the AI data. Setting information can be obtained. The AI setting unit 238 may determine a mapping relationship between the image related information and the DNN setting information in advance, and obtain DNN setting information mapped to the first image 1325 and 1345 related information.

Hereinafter, an AI encoding apparatus 1500 for AI encoding of an original image 1305 will be described with reference to FIG. 15 .

15 is a block diagram illustrating a configuration of an AI encoding apparatus 1500 according to an embodiment.

Referring to FIG. 15 , the AI encoding apparatus 1500 may include an AI encoding unit 1510 and a transmission unit 1530 . The AI encoding unit 1510 includes the original image resolution determining unit 1511, the AI 1:1 preprocessing unit 1512, the AI downscaling unit 1513, the first encoding unit 1514, the data processing unit 1516, and the AI setting. part 1518 . 15 illustrates the AI encoder 1510 and the transmitter 1530 as separate devices, the AI encoder 1510 and the transmitter 1530 may be implemented through one processor. In this case, it may be implemented as a dedicated processor, or it may be implemented through a combination of an AP or a general-purpose processor such as a CPU or GPU and S/W. In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

Also, the AI encoder 1510 and the transmitter 1530 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU.

In an embodiment, the first encoder 1510 is configured as a first processor, and an original image resolution determiner 1411, an AI one-to-one preprocessor 1412, an AI downscaler 1413, and a data processor ( 1516) and the AI setting unit 1518 may be implemented with a second processor different from the first processor, and the transmitter 1530 may be implemented with the first processor and a third processor different from the second processor.

The AI encoder 1510 determines the resolution of the original image 1305, and if the resolution is greater than a predetermined value, the AI downscale 1320 and the first image 1345 of the original image 1335 having a higher resolution. 1 encoding is performed, and AI-encoded data is transmitted to the transmitter 1530 . The transmitter 1530 transmits the AI encoded data to the AI decoding apparatus 200 . In addition, if the resolution of the original image 1305 is less than or equal to a predetermined value, AI 1-to-1 pre-processing 1310 of the original image 1315 having a small resolution and first encoding of the first image 1325 are performed, and AI encoding is performed. The data is transferred to the transmitter 1530 . The transmitter 1530 transmits the AI encoded data to the AI decoding apparatus 200 .

The image data includes data obtained as a result of the first encoding of the first images 1325 and 1345 . The image data is data obtained based on pixel values in the first images 1325 and 1345 , for example, a residual that is a difference between the first images 1325 and 1345 and prediction data of the first images 1325 and 1345 . It may contain data. Also, the image data includes information used in the first encoding process of the first images 1325 and 1345 . For example, the image data includes prediction mode information used to first encode the first images 1325 and 1345, motion information, and quantization parameter related information used to first encode the first images 1325 and 1345, etc. may include

AI data is information that enables the AI upscaling unit 236 to AI upscale the second image 1355 to an upscale target corresponding to the downscale target of the first DNN or one-to-one of the third DNN. It includes information enabling AI to upscale the second image 1355 to an upscale target corresponding to the preprocessing target. In one example, the AI data may include difference information between the original image 1335 and the AI downscaled first image 1345 . In an example, the AI data may include resolution information of the first image 1325 that has been pre-processed by AI one-to-one. Also, the AI data may include information related to the first images 1325 and 1345 . The information related to the first images 1325 and 1345 includes the resolution of the first images 1325 and 1345, the bitrate of image data obtained as a result of the first encoding of the first images 1225 and 1345, and the first image 1325, 1345) may include information on at least one of the codec types used in the first encoding.

In an embodiment, the AI data may include an identifier of mutually agreed DNN configuration information so that the second image 1355 can be AI upscaled to an upscale target corresponding to the downscale target of the first DNN. .

In one embodiment, the AI data may include an identifier of mutually agreed DNN configuration information so that the second image 1355 can be AI upscaled to an upscale target corresponding to the one-to-one preprocessing target of the third DNN. can

Also, in an embodiment, the AI data may include DNN configuration information settable in the second DNN.

The AI downscaler 1513 may acquire the AI downscaled first image 1345 from the original image 1335 through the first DNN. The AI downscaling unit 1513 may AI downscale the original image 1335 by using the DNN setting information provided from the AI setting unit 1518 . The AI setting unit 1518 may determine a downscale target of the original image 1235 based on a predetermined criterion.

In order to acquire the first image 1345 matching the downscale target, the AI setting unit 1518 may store a plurality of DNN setting information settable in the first DNN. The AI setting unit 1518 obtains DNN setting information corresponding to a downscale target among a plurality of DNN setting information, and provides the obtained DNN setting information to the AI downscale unit 1513 .

Each of the plurality of pieces of DNN configuration information may be trained to acquire the first image 1345 having a predetermined resolution and/or a predetermined image quality. For example, the DNN setting information of any one of the plurality of DNN setting information is the first image 1345 having a resolution that is 1/2 times smaller than the resolution of the original image 1335, for example, 4K (4096 * 2160). may include information for obtaining the first image 1345 of 2K (2048 * 1080) that is 1/2 times smaller than the original image 1335 of A first image 1345 with a resolution of 1/4 times smaller than that, for example, a first image 1345 of 2K (2048*1080) which is 1/4 times smaller than an original image 1335 of 8K (8192*4320) ) may include information for obtaining

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 1518 may obtain DNN setting information obtained by combining some selected among the lookup table values according to the downscale target, and may provide the obtained DNN setting information to the AI downscaling unit 1513 .

According to an embodiment, the AI setting unit 1518 may determine the structure of the DNN corresponding to the downscale target, and obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel.

The plurality of DNN configuration information for AI downscaling of the original image 1335 may have an optimized value by performing joint training between the first DNN and the second DNN. Here, each DNN configuration information includes at least one of the number of convolutional layers included in the first DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel.

The AI downscaling unit 1513 sets the first DNN as the DNN setting information determined for the AI downscaling of the original image 1335, and sets the first DNN to the first image 1345 of a predetermined resolution and/or a predetermined quality. can be obtained through When DNN setting information for AI downscaling of the original image 1335 is obtained among a plurality of DNN setting information, each layer in the first DNN may process input data based on information included in the DNN setting information. .

Hereinafter, a method for the AI setting unit 1518 to determine the downscale target will be described. The downscale target may indicate, for example, how much the resolution of the first image 1345 with reduced resolution should be obtained from the original image 1335 .

The AI setting unit 1518 acquires one or more pieces of input information. In an embodiment, the input information includes a target resolution of the first image 1345, a target bitrate of the image data, a bitrate type of the image data (eg, a variable bitrate type, a constant bitrate type, or an average bitrate type); Color format to which AI downscale is applied (luminance component, chrominance component, red component, green component, or blue component, etc.), codec type for the first encoding of the first image 1345, compression history information, and original image 1305 may include at least one of a resolution of , and a type of the original image 1305 .

One or more pieces of input information may be stored in advance in the AI encoding apparatus 1500 or may include information received from a user.

The AI setting unit 1518 controls the operation of the AI downscale unit 1513 based on the input information. In an embodiment, the AI setting unit 1518 may determine a downscale target according to input information, and provide DNN setting information corresponding to the determined downscale target to the AI downscale unit 1513 .

In an embodiment, the AI setting unit 1518 transmits at least a portion of the input information to the first encoder 1514 so that the first encoder 1514 sets a bitrate of a specific value, a bitrate of a specific type, and a specific codec. may cause the first image 1345 to be first encoded.

In an embodiment, the AI setting unit 1518 sets the compression rate (eg, a difference in resolution between the original image 1335 and the first image 1345 , the target bitrate), and the compression quality (eg, the bitrate type). ), compression history information, and a downscale target may be determined based on at least one of the type of the original image 1305 .

In an example, the AI setting unit 1518 may determine a downscale target based on a preset or a compression rate or compression quality input from a user.

As another example, the AI setting unit 1518 may determine the downscale target by using the compression history information stored in the AI encoding apparatus 1500 . For example, according to the compression history information available to the AI encoding apparatus 1500, the encoding quality or compression rate preferred by the user may be determined, and the downscale target may be determined according to the encoding quality determined based on the compression history information. can For example, the resolution and quality of the first image 1345 may be determined according to the encoding quality that has been most frequently used according to the compression history information.

As another example, the AI setting unit 1518 sets the encoding quality that has been used more than a predetermined threshold value (eg, average quality of encoding quality that has been used more than a predetermined threshold value) according to the compression history information. A downscale target may be determined based on it.

As another example, the AI setting unit 1518 may determine the downscale target based on the resolution and type (eg, file format) of the original image 1335 .

In one embodiment, when the original image 1335 is composed of a plurality of frames, the AI setting unit 1518 independently acquires DNN setting information for each predetermined number of frames, and uses the independently acquired DNN setting information to AI downscaling. It may be provided as part 1513 .

In one example, the AI setting unit 1518 may classify frames constituting the original image 1335 into a predetermined number of groups, and independently obtain DNN setting information for each group. The same or different DNN configuration information may be obtained for each group. The number of frames included in the groups may be the same or different for each group.

In another example, the AI setting unit 1518 may independently determine DNN setting information for each frame constituting the original image 1335 . The same or different DNN configuration information may be obtained for each frame.

The AI 1-to-1 preprocessor 1512 may acquire the AI 1-to-1 preprocessed first image 1325 from the original image 1315 through the third DNN. The AI one-to-one preprocessor 1512 may perform AI one-to-one preprocessing of the original image 1315 using the DNN setting information provided from the AI setting unit 1518 . The AI setting unit 1518 may determine a one-to-one preprocessing target of the original image 1315 based on a predetermined criterion.

In order to obtain the first image 1325 matching the one-to-one preprocessing target, the AI setting unit 1518 may store a plurality of DNN setting information settable in the third DNN. The AI setting unit 1518 obtains DNN setting information corresponding to a one-to-one pre-processing target among a plurality of DNN setting information, and provides the obtained DNN setting information to the AI one-to-one pre-processing unit 1512 .

Each of the plurality of DNN configuration information may be trained to acquire a first image 1325 in which detailed features of the original image are well preserved and/or a first image 1325 having a predetermined image quality. For example, the DNN setting information of any one of the plurality of DNN setting information is the first image 1325 in which the detailed characteristics of the image are well preserved while maintaining the resolution of the original image 1315, for example, 2K ( Although the resolution of the original image 1315 of 2048*1080 is the same, information for acquiring the first image 1325 in which detailed characteristics of the original image 1315 are well preserved may be included. Here, the detailed feature of the original image may be a part with high spatiotemporal complexity. The other DNN setting information is the first image 1325 with improved image quality while maintaining the resolution of the original image 1315 as it is, for example, the same resolution as the original image 1315 of 2K (2048 * 1080), but the quality Information for acquiring the improved first image 1325 may be included.

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 1518 may provide the AI one-to-one preprocessing unit 1512 with DNN setting information obtained by combining some selected among the lookup table values according to the one-to-one preprocessing target.

Depending on the implementation, the AI setting unit 1518 may determine the structure of the DNN corresponding to the one-to-one preprocessing target, and obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel. have.

A plurality of DNN setting information for AI one-to-one preprocessing of the original image 1315 is obtained by fixing the DNN setting information of the second DNN after the first DNN and the second DNN are jointly trained so that the third DNN and the second DNN are connected. By joint training, it is possible to have an optimized value. Here, each DNN configuration information includes at least one of the number of convolution layers included in the third DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel.

The AI one-to-one pre-processing unit 1512 sets the third DNN as the DNN setting information determined for the AI one-to-one pre-processing of the original image 1315 , and the first image in which detailed features of the original image 1315 are preserved. 1325 and/or the first image 1325 having a predetermined quality may be acquired through the third DNN. When DNN setting information for AI one-to-one pre-processing of the original image 1315 is obtained among a plurality of DNN setting information, each layer in the third DNN processes the input data based on the information included in the DNN setting information. can

Hereinafter, a method for the AI setting unit 1518 to determine a one-to-one pre-processing target will be described. The one-to-one preprocessing target determines, for example, how much the first image 1325 that preserves the detailed features of the original image while maintaining the resolution of the original image 1315 should be acquired or the resolution of the original image 1315 . It may indicate how much the first image 1325 with improved image quality should be acquired while maintaining it.

The AI setting unit 1518 acquires one or more pieces of input information. In an embodiment, the input information includes a target resolution of the first image 1325, a target bitrate of the image data, a bitrate type of the image data (eg, a variable bitrate type, a constant bitrate type, or an average bitrate type); Color format to which AI downscale is applied (luminance component, chrominance component, red component, green component, or blue component, etc.), codec type for the first encoding of the first image 1325, compression history information, and the original image 1305 may include at least one of a resolution of , and a type of the original image 1305 .

One or more pieces of input information may be stored in advance in the AI encoding apparatus 1500 or may include information received from a user.

The AI setting unit 1518 controls the operation of the AI one-to-one preprocessor 1512 based on the input information. In an embodiment, the AI setting unit 1518 determines a one-to-one pre-processing target according to the input information, and provides DNN setting information corresponding to the determined one-to-one pre-processing target to the AI one-to-one pre-processing unit 1512 . can

In an embodiment, the AI setting unit 1518 transmits at least a portion of the input information to the first encoder 1514 so that the first encoder 1514 sets a bitrate of a specific value, a bitrate of a specific type, and a specific codec. may cause the first image 1325 to be first encoded.

In one embodiment, the AI setting unit 1518 is at least one of the compression rate (eg, target bitrate), compression quality (eg, bitrate type), compression history information, and the type of the original image 1315 . Based on the one-to-one pre-processing target can be determined.

In one example, the AI setting unit 1518 may determine a one-to-one preprocessing target based on a preset or a compression rate or compression quality input from a user.

As another example, the AI setting unit 1518 may determine a one-to-one preprocessing target using compression history information stored in the AI encoding apparatus 1500 . For example, according to the compression history information available to the AI encoding apparatus 1500, the encoding quality or compression ratio preferred by the user may be determined, and the one-to-one preprocessing target may be determined based on the encoding quality determined based on the compression history information. This can be determined. For example, the resolution and quality of the first image 1325 may be determined according to the encoding quality that has been most frequently used according to the compression history information.

As another example, the AI setting unit 1518 sets the encoding quality that has been used more than a predetermined threshold value (eg, the average quality of encoding qualities that have been used more than a predetermined threshold value) according to the compression history information. Based on the one-to-one pre-processing target may be determined.

As another example, the AI setting unit 1518 may determine a one-to-one preprocessing target based on the resolution and type (eg, file format) of the original image 1315 .

In one embodiment, when the original image 1215 is composed of a plurality of frames, the AI one-to-one setting unit 1518 independently obtains DNN setting information for each predetermined number of frames, and uses the independently acquired DNN setting information. It may be provided to the AI one-to-one preprocessor 1512 .

In one example, the AI setting unit 1518 may classify frames constituting the original image 1315 into a predetermined number of groups, and independently obtain DNN setting information for each group. The same or different DNN configuration information may be obtained for each group. The number of frames included in the groups may be the same or different for each group.

In another example, the AI setting unit 1518 may independently determine DNN setting information for each frame constituting the original image 1315 . The same or different DNN configuration information may be obtained for each frame.

Hereinafter, an exemplary structure of the third DNN 1610, which is the basis for AI one-to-one preprocessing, will be described.

14 is an exemplary diagram illustrating a third DNN 1610 for AI one-to-one preprocessing of an original image 1315 .

14 , the original image 1315 is input to the first convolutional layer 1410 . The first convolutional layer 1410 performs convolution processing on the original image 1315 using 16 filter kernels having a size of 3×3. The 16 feature maps generated as a result of the convolution process are input to the first activation layer 1430 . The first activation layer 1420 may provide non-linear characteristics to 16 feature maps.

The first activation layer 1420 determines whether to transfer sample values of the feature maps output from the first convolution layer 1410 to the second convolution layer 1430 . For example, some sample values among the sample values of the feature maps are activated by the first activation layer 1420 and transferred to the second convolution layer 1430 , and some sample values are activated by the first activation layer 1420 . It is deactivated and is not transferred to the second convolutional layer 1430 . Information indicated by the feature maps output from the first convolutional layer 1410 is emphasized by the first activation layer 1420 .

An output 1425 of the first activation layer 1420 is input to the second convolutional layer 1430 . The second convolution layer 1430 performs convolution processing on input data using 16 filter kernels having a size of 3×3. The 16 feature maps output as a result of the convolution process are input to the second activation layer 1440 , and the second activation layer 1440 may provide non-linear characteristics to the 16 feature maps.

The output 1445 of the second activation layer 1440 is input to the third convolution layer 1450 . The third convolution layer 1450 performs convolution processing on input data using one filter kernel having a size of 3×3. As a result of the convolution process, one image may be output from the third convolutional layer 1450 . The third convolutional layer 1450 is a layer for outputting a final image and obtains one output by using one filter kernel. Finally, the image output from the third convolution layer 1450 and the original image 1315 that is the input image are added to obtain the final output image 1325 of the third DNN. According to the example of the present disclosure, the original image 1315 is added to the convolution operation result of the third convolutional layer 1450 to have the same resolution as the original image 1315 , and detailed features of the original image 1315 are preserved. A first image 1325 may be output.

DNN setting information indicating the number of filter kernels of the first convolutional layer 1410, the second convolutional layer 1430, and the third convolutional layer 1450 of the third DNN 1610, the parameters of the filter kernel, etc. There may be a plurality, and the plurality of DNN configuration information should be associated with the plurality of DNN configuration information of the second DNN. The association between the plurality of DNN setting information of the third DNN and the plurality of DNN setting information of the second DNN is performed using the DNN setting information of the second DNN obtained after the association learning of the first DNN and the second DNN. It can be implemented through joint learning of the DNN and the second DNN.

14 shows that the third DNN 1610 includes three convolutional layers 1410 , 1430 , 1450 and two activation layers 1420 , 1440 , but this is only an example, and the implementation Accordingly, the number of convolutional layers and activation layers may be variously changed. Also, according to an embodiment, the third DNN 1610 may be implemented through a recurrent neural network (RNN). In this case, it means changing the CNN structure of the third DNN 1610 according to the example of the present disclosure to the RNN structure.

In an embodiment, the AI 1-to-1 preprocessor 1512 may include at least one ALU for a convolution operation and an activation layer operation. The ALU may be implemented as a processor. For the convolution operation, the ALU may include a multiplier that performs a multiplication operation between the sample values of the feature map output from the original image 1315 or the previous layer and the sample values of the filter kernel, and an adder that adds the result values of the multiplication. have. In addition, for the calculation of the activation layer, the ALU is a multiplier that multiplies an input sample value by a weight used in a predetermined sigmoid function, a Tanh function, or a ReLU function, and compares the multiplication result with a predetermined value to calculate the input sample value It may include a comparator that determines whether to transfer to the next layer.

Referring back to FIG. 15 , the AI setting unit 1518 transmits AI data to the data processing unit 1516 . AI data allows the AI upscale unit 236 to AI upscale the second image 1355 to an upscale target corresponding to the downscale target of the first DNN or the one-to-one preprocessing target of the third DNN. include information. The first encoder 1514 receiving the one-to-one pre-processed first image 1325 from the AI 1:1 preprocessor 1512 or the downscaled first image 1345 from the AI downscaler 1513 is The amount of information included in the first images 1325 and 1345 may be reduced by first encoding the first images 1325 and 1345 according to a frequency transformation-based image compression method. As a result of the first encoding through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1), image data is obtained. The image data is obtained according to a rule of a predetermined codec, that is, a syntax. For example, the image data may include residual data that is a difference between the first images 1325 and 1345 and the prediction data of the first images 1325 and 1345 , and prediction used to first encode the first images 1325 and 1345 . It may include mode information, motion information, and information related to a quantization parameter used for first encoding the first images 1325 and 1345, and the like. The image data obtained as a result of the first encoding by the first encoder 1514 is provided to the data processor 1516 .

The data processing unit 1516 generates AI encoded data including the image data received from the first encoding unit 1514 and the AI data and image data received from the AI setting unit 1518 . In an embodiment, the data processing unit 1516 may generate AI-encoded data including image data and AI data in a separate state. As an example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream.

In another embodiment, the data processing unit 1516 may include AI data in image data obtained as a result of the first encoding by the first encoder 1514 and generate AI encoded data including the corresponding image data. . For example, the data processing unit 1516 may generate image data in the form of one bitstream by combining a bitstream corresponding to image data and a bitstream corresponding to AI data. To this end, the data processing unit 716 may express AI data as bits having a value of 0 or 1, that is, a bitstream. In an embodiment, the data processing unit 1516 may include a bitstream corresponding to AI data in supplemental enhancement information (SEI), which is an additional information area of a bitstream obtained as a result of the first encoding.

AI-encoded data is transmitted to the transmitter 1530 . The transmitter 1530 transmits the AI-encoded data obtained as a result of the AI-encoding through a network.

In one embodiment, the AI encoded data is a hard disk, a magnetic medium such as a floppy disk and magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk. medium) may be stored in a data storage medium including

Hereinafter, with reference to FIG. 16 , a method of jointly training the first DNN 800 and the second DNN 300 and then jointly training the third DNN 1610 with the second DNN 300 will be described. .

16 is a diagram for explaining a method of training the third DNN 1610 in association with the second DNN 300 .

In an embodiment, the AI-encoded original image 1305 through the AI encoding process is restored or upscaled to the third image 1365 through the AI decoding process. The third image 1365 and the original image obtained as a result of the AI decoding In order to maintain the similarity to (1305), correlation is required between the AI encoding process and the AI decoding process. That is, information lost in the AI encoding process should be able to be restored in the AI decoding process. 300) is required.

16 is a diagram for explaining a method of joint training 1600 of the third DNN 1610 and the second DNN 300 that is performed after the joint training of the first DNN 800 and the second DNN 300 of FIG. 11 described above. As a diagram for this purpose, the cooperative training of the first DNN 800 and the second DNN 300 other than the cooperative training 1600 of the third DNN 1610 and the second DNN 300 is the same as described above in FIG. 11 , omitted here.

For accurate AI decoding, it is ultimately necessary to reduce the low-resolution loss information 1620 corresponding to the comparison result between the original training image 1101 and the fourth training image 1602 shown in FIG. 16 . Since the joint training of the third DNN and the second DNN is performed using the parameters of the second DNN obtained from the first and second DNN joint training performed first, the low-resolution loss information 1620 is the third DNN ( 1610).

First, the training process shown in FIG. 16 will be described.

In FIG. 16 , the original training image is an image to be compared with an AI upscaled image after AI one-to-one pre-processing to preserve the detailed features of a low-resolution image, and the reduced training image 1601 is an AI one-to-one pre-processing It is an image corresponding to an image having a small resolution as a target of , and the fourth training image 1602 is an AI upscaled image after the reduced training image 1601 is pre-processed by AI one-to-one.

The original training image 1101 includes a still image or a moving image composed of a plurality of frames. In an embodiment, the original training image 1101 may include a still image or a luminance image extracted from a moving picture including a plurality of frames. Also, according to an embodiment, the original training image 1101 may include a still image or a patch image extracted from a moving image including a plurality of frames. When the original training image 1101 includes a plurality of frames, the reduced training image 1601 and the fourth training image 1602 also include a plurality of frames. When a plurality of frames of the reduced training image 1601 are sequentially input to the third DNN 1610 , the plurality of frames of the fourth training image 1602 are transmitted through the third DNN 1610 and the second DNN 300 . can be obtained sequentially.

For joint training of the third DNN 1610 and the second DNN 300 , a reduced training image 1601 is input to the third DNN 1610 . The reduced training image 1601 input to the third DNN 1610 is pre-processed by AI one-to-one, and then input to the second DNN 300 to output a fourth training image 1602 as a result of AI upscaling.

Referring to FIG. 16 , the AI one-to-one pre-processing result is being input to the second DNN 300. According to an embodiment, the result obtained through the first encoding and first decoding of the AI one-to-one pre-processing result is It may be input to the second DNN 300 . MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9 and Any one of AV1 codecs may be used. Specifically, for the first encoding and the first decoding, any one of MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1 may be used.

Referring to FIG. 16 , a reduced training image 1601 is acquired by legacy downscaling from an original training image 1101 . Here, the legacy downscale may include at least one of a bilinear scale, a bicubic scale, a lanczos scale, and a stair step scale.

In order to enable upscaling while maintaining the structural features of the low-resolution original image, training is performed using the reduced-resolution training image while maintaining the structural features of the original training image.

Before the training proceeds, the third DNN 1610 may be set with predetermined DNN configuration information, and the second DNN may be set with DNN configuration information obtained as a result of the joint training of the first DNN and the second DNN previously performed. have. As training progresses, low-resolution loss information 1620 may be determined.

The low-resolution loss information 1620 may be determined based on a comparison result between the original training image 1101 and the fourth training image 1602 . In one example, the fourth training image obtained by upscaling the one-to-one preprocessing result that preserves the detailed features of the reduced training image 1601 with a small resolution and the reduced training image and the structural features correspond to the difference between the original training image can

The low-resolution loss information 1620 includes an L1-norm value, an L2-norm value, a Structural Similarity (SSIM) value, and a Peak Signal-To (PSNR-HVS) value for the difference between the original training image 1101 and the fourth training image 1602 . It may include at least one of a Noise Ratio-Human Vision System) value, a Multiscale SSIM (MS-SSIM) value, a Variance Inflation Factor (VIF) value, and a Video Multimethod Assessment Fusion (VMAF) value. The low-resolution loss information 1620 indicates how similar the fourth training image 1602 is to the original training image 1101 . As the low-resolution loss information 1620 is smaller, the fourth training image 1602 is more similar to the original training image 1101 .

Referring to FIG. 16 , low-resolution loss information 1620 is used for training of the third DNN 1610 .

The third DNN 1610 may update a parameter such that the final loss information determined based on the low-resolution loss information 1620 is reduced or minimized. In the second DNN 300 , parameters determined in the joint training of the first DNN and the second DNN previously performed are used as they are.

Final loss information for training of the third DNN may be determined as in Equation 2 below.

[Equation 2]

LossLU = e*low resolution loss information

In Equation 2, LossLU represents final loss information to be reduced or minimized for training of the third DNN 1610 . Also, e may correspond to a predetermined weight.

After the joint training of the first DNN and the second DNN, the third DNN 1610 updates the parameters in a direction in which the LossLU of Equation 2 is decreased while using the parameters of the second DNN as it is.

Since the joint training of the third DNN and the second DNN is performed after the joint training of the first DNN and the second DNN, as a whole, referring to FIGS. 11 and 16 , the third DNN 1610 according to the LossLU derived from the training process. When the parameters of are updated, the fourth training image 1602 obtained based on the parameters of the second DNN updated according to the joint training of the first DNN and the second DNN and the parameters of the third DNN is displayed in the previous training process. is different from the fourth training image 1602, and when the fourth training image 1602 is different from the fourth training image 1602 in the previous training process, the low resolution loss information 1620 is also newly determined, Accordingly, the third DNN 1610 updates the parameters. The third DNN can optimize the parameters of the third DNN so that the low-resolution loss information is minimized by using the parameters updated according to the results of the joint training of the first DNN and the second DNN performed before the joint training with the second DNN. have.

Previously, it has been described that the AI setting unit 238 of the AI decoding apparatus 200 and the AI setting unit 1518 of the AI encoding apparatus 1500 store a plurality of DNN setting information, the AI setting unit 238 and the AI The method of training each of the plurality of DNN configuration information for the first DNN and the second DNN stored in the setting unit 1518 is as described above with reference to FIG. 11 , the plurality of DNN setting information stored in the AI setting unit 1518 is Explain how to train.

As described in relation to Equation 2, in the case of the third DNN 1601, the degree of similarity between the structural information of the original training image 1101 and the structural information of the fourth training image 1602 (low-resolution loss information 1620) The parameters are updated taking into account.

In more detail, the parameters of the third DNN may be updated so that the original training image and the fourth training image having similar structural features to the reduced training image are similar. Accordingly, an upscale image in which the structural features of the reduced training image having a small resolution are well preserved may be generated.

17 is a diagram for explaining a training process of the third DNN 1610 and the second DNN 300 by the training apparatus 1200 .

Training of the third DNN 1610 and the second DNN 300 described with reference to FIG. 16 may be performed by the training apparatus 1000 . The training apparatus 1200 includes a first DNN 800 , a second DNN 300 , and a third DNN 1610 . The training device 1200 may be, for example, the AI encoding device 1500 or a separate server.

The training process of FIG. 17 is performed after the above-described training process of FIG. 12 is performed first.

Referring to FIG. 16 , the training apparatus 1200 initially sets DNN configuration information of the third DNN 1610 ( S1710 ), and loads the DNN configuration information of the second DNN obtained in the training process of FIG. 10 ( S1720 ). ). Accordingly, the third DNN 1610 and the second DNN 300 may operate according to predetermined DNN configuration information. The DNN setting information includes at least the number of convolution layers included in the third DNN 1610 and the second DNN 300 , the number of filter kernels for each convolutional layer, the size of filter kernels for each convolutional layer, and parameters of each filter kernel. It can contain information about one.

The training apparatus 1200 inputs the reduced training image 1601 to the third DNN 1610 (S1730). The reduced training image 1101 may include at least one frame constituting a still image or a moving image.

The third DNN 1610 processes the reduced training image 1601 according to the initially set DNN setting information, and outputs a result of AI 1-to-1 pre-processing from the reduced training image 1601 ( S1740 ). FIG. 16 shows that the result output from the third DNN 1610 is directly input to the second DNN 300 , but the result output from the third DNN 1610 is outputted from the third DNN 1610 by the training device 1200 . (300) may be input. Also, the training apparatus 1200 may input a second image obtained by first encoding and first decoding a result output from the third DNN 1610 using a predetermined codec to the second DNN 300 .

The second DNN 300 processes the result output from the third DNN 1610 according to the loaded DNN setting information, and an AI upscaled fourth training image 1620 from the result output from the third DNN 1610 . is output (S1750).

The training apparatus 1200 calculates low-resolution loss information 1620 by comparing the original training image 1101 with the fourth training image 1620 ( S1760 ).

The third DNN 1610 updates the initially set DNN configuration information through a back propagation process based on the low resolution loss information (S1770).

Thereafter, the training device 1200, the third DNN 1610, and the second DNN 300 set the DNN while repeating the processes of S1240 to S1285 of FIG. 12 and S1710 to S1770 of FIG. 17 until the low-resolution loss information is minimized. Update the information. At this time, during each iteration process, the third DNN 1610 and the second DNN 300 operate according to the DNN configuration information updated in the previous process.

18 is a flowchart illustrating an AI decoding method according to an embodiment.

In operation S1810, the AI decoding apparatus 200 receives AI encoded data including image data and AI data. The AI decoding apparatus 200 may receive AI encoded data from the AI encoding apparatus 1500 through a network. The AI decoding apparatus 200 may acquire AI encoded data stored in a data storage medium.

In operation S1820 , the AI decoding apparatus 200 acquires a second image 1355 based on the image data. Specifically, the AI decoding apparatus 200 reconstructs the second image 1355 corresponding to the first images 1325 and 1345 by decoding the image data based on the image restoration method using frequency transformation.

In step S1830 , the AI decoding apparatus 200 acquires DNN setting information for AI upscaling of the second image 1355 from among a plurality of pre-stored DNN setting information. Each of the plurality of DNN setting information is optimized in connection with each of a plurality of DNN setting information used for AI downscaling of the original image 1335 and each of a plurality of DNN setting information used for AI one-to-one preprocessing of the original image 1315 Therefore, the second image 1355 can be AI upscaled according to an upscale target that matches the downscale target of the original image 1335 or an upscale target that matches the one-to-one preprocessing target of the original image 1315. You must select the DNN configuration information that is present.

In operation S1840, the AI decoding apparatus 200 generates an AI-upscaled third image 1365 from the second image 1355 through the second DNN operating with the DNN setting information obtained in operation S1830. The third image 1365 may be output from the AI decoding apparatus 200 and displayed through the display apparatus, or may be displayed after being post-processed.

When the DNN configuration information is preset in the second DNN, the AI decoding apparatus 200 sets the second DNN to the selected parameter set when the DNN configuration information selected in step S1830 is different from the preset DNN configuration information. .

19 is a flowchart illustrating an AI encoding method according to an embodiment.

In step S1910 , the AI encoding apparatus 1500 determines the resolution size of the original image 1305 .

In step S1920, when the resolution size of the original image 1305 is greater than a predetermined value, the AI encoding apparatus 1500 obtains the AI downscaled first image 1345 from the original image 1335 through the first DNN. .

The AI encoding apparatus 1500 may determine a downscale target based on a predetermined criterion and obtain DNN setting information corresponding to the downscale target from among a plurality of pre-stored DNN setting information. In addition, the AI encoding apparatus 1500 may AI downscale the original image 1335 through the first DNN operating according to the obtained DNN setting information.

In step S1930, when the resolution size of the original image 1305 is less than or equal to a predetermined value, the AI encoding apparatus 1500 performs AI 1-to-1 preprocessing of the first image 1325 from the original image 1315 through the third DNN. acquire

The AI encoding apparatus 1500 may determine a one-to-one pre-processing target based on a predetermined criterion, and obtain DNN setting information corresponding to the one-to-one pre-processing target among a plurality of pre-stored DNN setting information. In addition, the AI encoding apparatus 1500 may pre-process the original image 1315 AI one-to-one through the third DNN operating according to the obtained DNN setting information.

In operation S1940, the AI encoding apparatus 1500 first encodes the downscaled first image 1345 or the one-to-one preprocessed first image 1325 to generate image data. Specifically, the AI encoding apparatus 1500 generates image data corresponding to the first images 1325 and 1345 by encoding the first images 1325 and 1345 based on the image compression method using frequency transformation.

In step S1950 , the AI encoding apparatus 1500 transmits AI encoded data including AI data and image data including information related to AI one-to-one preprocessing or AI downscaling. The AI data includes information for selecting DNN setting information of the second DNN for AI upscaling of the second image 1355 . In an embodiment, the AI-encoded data may be stored in a data storage medium.

As described above, since the third DNN and the second DNN are jointly trained using the DNN setting information of the second DNN obtained after the first DNN and the second DNN are jointly trained, the AI encoding device 1400 is specific When the original image 1335 with a large resolution is downscaled by AI as the downscale target of , when the AI encoding apparatus 1500 pre-processes the original image 1315 with a small resolution as a specific one-to-one pre-processing target, the AI decoding apparatus 200 also corresponds to the one-to-one pre-processing target. The second image 1355 should be AI upscaled as a scale target.

Accordingly, the AI data is information or resolution that enables the AI decoding apparatus 200 to AI upscale the second image 1355 to an upscale target corresponding to the downscale target of the original image 1335 having high resolution. includes information enabling AI to upscale the second image 1355 to an upscale target corresponding to the one-to-one pre-processing target of the small original image 1315 . Specifically, the AI data includes information used to obtain DNN configuration information corresponding to the upscale target.

The AI decoding device 200 that has received the AI data can infer or know which DNN setting information the AI encoding device 1500 AI downscaled the original image 1335 with a high resolution, and accordingly, the AI downscale It is possible to acquire DNN configuration information corresponding to the used DNN configuration information, and use the obtained DNN configuration information to upscale AI.

In addition, the AI decoding device 200 that has received the AI data can infer or know with what DNN setting information the AI encoding device 1500 pre-processed the original image 1315 with a small resolution one-to-one AI, and accordingly It is possible to acquire DNN configuration information corresponding to the DNN configuration information used for AI one-to-one preprocessing, and use the obtained DNN configuration information to upscale AI.

20 is a diagram for explaining an artificial intelligence (AI) encoding process and an AI decoding process according to an embodiment.

As shown in FIG. 20 , according to an exemplary embodiment of the present disclosure, the original image 2005 is pre-processed by AI one-to-one pre-processing (2010) to obtain the first image 2015, which has been pre-processed one-to-one. And, since the first encoding 2020 and the first decoding 2030 are performed on the first image 2015 in which the specific characteristics of the original image 2005 are emphasized, information that is difficult to restore when the original image is upscaled can be compensated for When the image is upscaled, if the complexity of the image is high, information may be lost and image quality may be deteriorated. Therefore, it is necessary to preserve the detailed features of the original image through one-to-one preprocessing to increase the memorization at the time of upscaling.

Referring to FIG. 20, in an embodiment, in the AI encoding process, AI 1-to-1 pre-processing (2010) the original image 2005 to obtain a 1-to-1 pre-processed first image 2015, and The image is first encoded (2020). In the AI decoding process, AI encoding data including AI data and image data obtained as a result of AI encoding is received, a second image 2035 is obtained through a first decoding 2030, and a second image 2035 is obtained. A third image 2045 is acquired by AI upscaling 2040 .

Looking at the AI encoding process in more detail, AI one-to-one pre-processing (2010) of the original image 2005 is performed to obtain the first image 2015 in which the detailed features of the original image 2005 are preserved. At this time, AI one-to-one pre-processing is performed based on AI, and the AI for AI one-to-one pre-processing must be jointly trained in connection with AI for AI upscaling (2040). Because, when AI for AI one-to-one preprocessing (2010) and AI for AI upscaling 2040 are trained separately, the third image 2045 upscaled through AI decoding is the original image ( 2005) may not be included. In this case, the parameters of the AI for the AI upscaling 2040 are trained in conjunction with the AI for the AI downscale 110 of FIG. 1 or the AI for the AI downscale 1320 of FIG. The parameters are used as-is. That is, when the parameters of AI for AI one-to-one preprocessing are updated during the joint training of AI for AI (2010) and AI upscaling (2040) for AI one-to-one preprocessing, AI upscaling (2040) The parameters of the AI are fixed and not updated. Specifically, AI for AI upscaling (2040) and AI for AI downscaling (110, 1320) are firstly trained in conjunction, and AI parameters for AI upscaling (2040) obtained as a result of training are used as it is. AI for one-to-one pre-processing (2010) and AI for AI upscaling (2040) are trained in conjunction, and parameters of AI for AI one-to-one pre-processing (2010) are obtained. The parameters of AI for the AI upscaling (2040) are trained in conjunction with the AI for the AI downscaling (110, 1320) and using the acquired parameters as they are, while maintaining the performance of the AI for the AI upscaling (2040), When upscaling, parameters of AI for AI one-to-one preprocessing 2010 that can reflect detailed characteristics of an image may be acquired. Accordingly, the original image may be upscaled without degradation of image quality.

In an embodiment of the present disclosure, AI data may be used in order to maintain this linkage in the AI encoding process and the AI decoding process. Accordingly, the AI data obtained through the AI encoding process should include information indicating the upscale target, and in the AI decoding process, the second image 2035 is AI upscaled ( 2040) should be done.

AI for AI one-to-one preprocessing (2010), AI for AI downscaling (110, 1220), and AI for AI upscaling (2040) may be implemented as a deep neural network (DNN). As described above with reference to FIG. 16 , the first DNN and the second DNN are jointly trained through sharing of loss information under a predetermined target, and the third DNN is trained after the first DNN and the second DNN are first trained, and then the second Using the DNN configuration information of the DNN as it is, training is performed in conjunction with the second DNN. Since the third DNN and the second DNN are jointly trained under a predetermined target using the second DNN of the DNN setting information obtained after the joint training of the first DNN and the second DNN, the AI encoding device connects the third DNN and the second DNN Target information used during training may be provided to the AI decoding device, and the AI decoding device may perform AI upscale 2040 to a resolution targeting the second image 2035 based on the received target information.

The first encoding 2020 and the first decoding 2030 shown in FIG. 20 will be described in detail. The first image 2015 obtained by AI one-to-one pre-processing (2010) from the original image 2005 is first encoded ( 2020), the amount of information can be reduced. The first encoding 2030 includes a process of predicting the first image 2015 to generate prediction data, a process of generating residual data corresponding to a difference between the first image 2015 and the prediction data, and a spatial domain component. It may include a process of transforming the residual data into a frequency domain component, a process of quantizing the residual data transformed into a frequency domain component, and a process of entropy encoding the quantized residual data. Such a first encoding process 2020 is MPEG-2, H.264 Advanced Video Coding (AVC), MPEG-4, High Efficiency Video Coding (HEVC), VC-1, VP8, VP9 and AV1 (AOMedia Video 1). It can be implemented through one of the image compression methods using frequency transformation, etc.

The second image 2035 corresponding to the first image 2015 may be restored through the first decoding 2030 of image data. The first decoding 2030 includes a process of generating quantized residual data by entropy-decoding image data, a process of inverse quantizing the quantized residual data, a process of transforming the residual data of a frequency domain component into a spatial domain component, a process of predicting data It may include a process of generating , and a process of reconstructing the second image 2035 using prediction data and residual data. The first decoding 2030 process is image compression using frequency conversion such as MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1 used in the first encoding 2020 process. It may be implemented through an image restoration method corresponding to one of the methods.

The AI encoded data obtained through the AI encoding process includes image data obtained as a result of the first encoding 2020 of the first image 2015 and AI data related to AI one-to-one preprocessing 2020 of the original image 2005. may include The image data may be used in the first decoding 2030 process, and the AI data may be used in the AI upscaling 2040 process.

The image data may be transmitted in the form of a bitstream. The image data may include data obtained based on pixel values in the first image 2015 , for example, residual data that is a difference between the first image 2015 and prediction data of the first image 2015 . . In addition, the image data includes information used in the first encoding 2020 of the first image 2015 . For example, the image data includes prediction mode information used for the first encoding 2020 of the first image 2015, motion information, and quantization parameter related information used in the first encoding 2020. can do. The image data is an image compression method used in the first encoding 1230 process among image compression methods using frequency conversion such as MPEG-2, H.264 AVC, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1. It may be generated according to a rule of, for example, a syntax.

AI data is used for AI upscaling 2040 based on the second DNN. As described above, after the first DNN and the second DNN are jointly trained, the third DNN and the second DNN are additionally jointly trained using the obtained DNN setting information of the second DNN as it is, so the AI data is 2 Contains information that enables accurate AI upscaling 2040 of the second image 2035 through DNN to be performed. In the AI decoding process, AI upscaling 2040 may be performed to a resolution and/or image quality targeting the second image 2035 based on AI data.

AI data may be transmitted together with image data in the form of a bitstream.

According to an embodiment, AI data may be transmitted separately from image data in the form of frames or packets.

Alternatively, according to an embodiment, AI data may be transmitted while being included in image data.

The image data and AI data may be transmitted through the same network or different networks.

The AI decoding apparatus for AI decoding according to FIG. 20 may have the same configuration as the AI decoding apparatus of FIG. 2 described above.

Referring to FIG. 2 , the AI decoding apparatus 200 according to an embodiment includes a receiving unit 210 and an AI decoding unit 230 .

The AI decoding unit 230 may include a parsing unit 232 , a first decoding unit 234 , an AI upscaling unit 236 , and an AI setting unit 238 .

Although the receiver 210 and the AI decoder 230 are illustrated as separate devices in FIG. 2 , the receiver 210 and the AI decoder 230 may be implemented through one processor. In this case, the receiving unit 210 and the AI decoding unit 230 may be implemented as a dedicated processor, and a general-purpose processor such as an application processor (AP), a central processing unit (CPU), or a graphic processing unit (GPU) and S/W It may be implemented through a combination of In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

The receiver 210 and the AI decoder 230 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU. In one embodiment, the receiving unit 210 is implemented as a first processor, the first decoding unit 234 is implemented as a second processor different from the first processor, the parsing unit 232, the AI upscaling unit 236 and the AI setting unit 238 may be implemented as a third processor different from the first processor and the second processor.

The receiving unit 210 receives AI encoded data obtained as a result of AI encoding. As an example, the AI-encoded data may be a video file having a file format such as mp4 or mov.

The receiver 210 may receive AI-encoded data transmitted through a network. The receiving unit 210 outputs the AI encoded data to the AI decoding unit 230 .

In one embodiment, the AI-encoded data is a hard disk, magnetic media such as floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks. It may be obtained from a data storage medium including an optical medium).

The parsing unit 232 parses the AI encoded data and transmits image data generated as a result of the first encoding of the first image 115 to the first decoder 234 , and transmits the AI data to the AI setting unit 238 . transmit

In an embodiment, the parsing unit 232 may parse the image data and the AI data separately included in the AI encoded data. The parsing unit 232 may read the header in the AI-encoded data to distinguish the AI data and the image data included in the AI-encoded data. In one example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream.

The structure of AI-encoded data including separated AI data and image data is omitted as described above with reference to FIG. 9 .

In another embodiment, the parsing unit 232 parses the image data from the AI encoded data, extracts the AI data from the image data, transmits the AI data to the AI setting unit 238, and first decodes the remaining image data. may be transmitted to the unit 234 . That is, AI data may be included in image data. For example, AI data may be included in supplemental enhancement information (SEI), which is an additional information area of a bitstream corresponding to image data. The structure of the AI-encoded data including the image data including the AI data has been described above with reference to FIG. 10, and thus will be omitted.

In another embodiment, the parsing unit 232 divides the bitstream corresponding to the image data into a bitstream to be processed by the first decoding unit 234 and a bitstream corresponding to AI data, and divides each divided bitstream It may output to the first decoding unit 234 and the AI setting unit 238 .

The parsing unit 232 obtains image data included in the AI encoded data through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1). It can also be confirmed that it is the image data that has been processed. In this case, the corresponding information may be transmitted to the first decoder 234 so that the image data can be processed by the identified codec.

The first decoder 234 reconstructs the second image 135 corresponding to the first image 115 based on the image data received from the parser 232 . The second image 135 obtained by the first decoder 234 is provided to the AI upscaler 236 .

According to an embodiment, first decoding-related information such as prediction mode information, motion information, and quantization parameter information may be provided from the first decoding unit 234 to the AI setting unit 238 . The first decoding related information may be used to obtain DNN configuration information.

The AI data provided to the AI setting unit 238 of the AI decoding apparatus for AI decoding according to FIG. 20 includes information enabling AI upscaling of the second image 2035 . In this case, the upscale target of the second image 2035 should correspond to the one-to-one preprocessing of the third DNN. Therefore, AI data should include information that can identify the one-to-one preprocessing target of the third DNN.

Specifically exemplifying the information included in the AI data, there are resolution information of the first image 2015 that has been pre-processed one-to-one from the original image 2005 and information related to the first image 2015 that has been pre-processed one-to-one.

Since the resolution information of the one-to-one preprocessed first image 2015 is the same as that of the original image 2005 , an upscale target may be determined according to a predetermined criterion according to the resolution of the first image 2015 . For example, if the resolution size of the first image pre-processed one-to-one is 2K, the upscale target may be determined to be 4K. Also, the resolution information of the first image 2015 that has been pre-processed one-to-one may be resolution difference information. Specifically, since the resolution of the one-to-one preprocessed first image 2015 and the original image 2005 is the same, the resolution difference information may indicate 0. In addition, since the resolution of the first image 2015 is known through the resolution of the reconstructed second image 2035 and the degree of resolution conversion can be confirmed through this, the resolution information may be expressed only with the resolution information of the original image 2005. may

And, similarly, the information related to the first image 2015, which has been pre-processed one-to-one, includes the resolution of the first image 2015, the bitrate of image data obtained as a result of the first encoding of the first image 2015, and the first image. (2015) may include information on at least one of the codec types used in the first encoding.

The AI setting unit 238 is configured to configure the second image based on at least one of resolution information (or difference information indicating that the resolution difference is 0) and information related to the first image 2015 included in the AI data. An upscale target of 2035 may be determined. The upscale target may indicate, for example, to what resolution the second image 2035 should be upscaled. When the upscaling target is determined, the AI upscaling unit 236 AI upscales the second image 2035 through the second DNN to obtain a third image 2045 corresponding to the upscaling target.

The AI upscaling process through the second DNN is omitted as described above with reference to FIGS. 3 and 4 .

Hereinafter, a method in which the AI setting unit 238 determines the upscaling target and the AI upscaling unit 236 AI upscales the second image 2035 according to the upscaling target will be described.

In an embodiment, the AI setting unit 238 may store a plurality of DNN setting information settable in the second DNN.

Here, the DNN configuration information may include information on at least one of the number of convolutional layers included in the second DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel. The plurality of DNN configuration information may each correspond to various upscale targets, and the second DNN may operate based on the DNN configuration information corresponding to a specific upscale target. The second DNN may have a different structure according to the DNN configuration information. For example, the second DNN may include three convolutional layers according to certain DNN configuration information, and the second DNN may include four convolutional layers according to other DNN configuration information.

In an embodiment, the DNN configuration information may include only parameters of a filter kernel used in the second DNN. In this case, the structure of the second DNN is not changed, but only the parameters of the internal filter kernel may be changed according to the DNN configuration information.

The AI setting unit 238 may acquire DNN setting information for AI upscaling of the second image 2035 among a plurality of DNN setting information. Each of the plurality of DNN setting information used here is information for obtaining the third image 2045 of a predetermined resolution and/or predetermined quality, and is trained in association with the first DNN and then trained in association with the third DNN. .

For example, the DNN setting information of any one of the plurality of DNN setting information is a third image 2045 having a resolution twice that of the second image 2035, for example, a second image of 2K (2048 * 1080). It may include information for acquiring the third image 2045 of 4K (4096 * 2160) that is twice as large as the second image 2035, and the other DNN setting information is higher than the resolution of the second image 2035 by 4 Information for obtaining a third image 2045 of twice the resolution, for example, a third image 2045 of 8K (8192 * 4320) that is four times larger than the second image 2035 of 2K (2048 * 1080) may include

Each of the plurality of DNN configuration information is created in association with the DNN configuration information of the first DNN, and then is associated with the DNN configuration information of the third DNN of the AI encoding apparatus 2100 . The AI setting unit 238 obtains one DNN setting information from among a plurality of DNN setting information according to the resolution information of the first image 2015 of the DNN setting information of the third DNN. To this end, the AI setting unit 238 should check the information of the third DNN. In order for the AI setting unit 238 to check the information of the third DNN, the AI decoding apparatus 200 according to an embodiment receives AI data including the information of the third DNN from the AI encoding apparatus 2100 .

In other words, the AI setting unit 238 uses the information received from the AI encoding device 2100 to check the information targeted by the DNN setting information of the third DNN used to obtain the first image 2015, and , it is possible to obtain the DNN configuration information of the second DNN trained in association with it.

When DNN setting information for AI upscaling of the second image 2035 is obtained among a plurality of DNN setting information, the DNN setting information is transmitted to the AI upscaling unit 236, and a second DNN operating according to the DNN setting information. Based on the input data may be processed.

For example, when any one DNN configuration information is obtained, the AI upscaling unit 236 includes the first convolutional layer 310 and the second convolutional layer 330 of the second DNN 300 shown in FIG. 3 . ) and the third convolutional layer 350, the number of filter kernels included in each layer and parameters of the filter kernels are set to values included in the obtained DNN configuration information.

Specifically, the AI upscaling unit 236 determines that the parameters of the filter kernel of 3 X 3 used in any one convolution layer of the second DNN shown in FIG. 4 are {1, 1, 1, 1, 1, 1 , 1, 1, 1}, and when there is a change in DNN configuration information, the parameters of the filter kernel are changed to {2, 2, 2, 2, 2, 2, 2, 2, 2, parameters included in the changed DNN configuration information. } can be replaced.

The AI setting unit 238 may acquire DNN setting information for upscaling the second image 2035 among a plurality of DNN setting information based on the information included in the AI data, which is used to obtain the DNN setting information. AI data will be explained in detail.

In an embodiment, the AI setting unit 238 upscales the second image 2035 among the plurality of DNN setting information based on the resolution information of the one-to-one preprocessed first image 2015 included in the AI data. It is possible to obtain DNN configuration information for For example, since the resolution of the one-to-one preprocessed first image 2015 is the same as that of the original image 2005, the upscale target may be determined according to a predetermined criterion according to the resolution information of the first image 2015. have. For example, if the resolution size of the first image pre-processed one-to-one is 2K, the upscale target may be determined to be 4K.

In another embodiment, the AI setting unit 238 is based on the information related to the first image 2015 included in the AI data, DNN setting information for AI upscaling the second image 2015 among a plurality of DNN setting information. can be obtained. The AI setting unit 238 may determine a mapping relationship between the image related information and the DNN setting information in advance, and obtain DNN setting information mapped to the first image 2015 related information.

Hereinafter, the AI encoding apparatus 2100 for AI encoding of the original image 2005 will be described with reference to FIG. 21 .

21 is a block diagram illustrating a configuration of an AI encoding apparatus 2100 according to an embodiment.

Referring to FIG. 21 , the AI encoding apparatus 2100 may include an AI encoding unit 2110 and a transmission unit 2130 . The AI encoder 2110 may include an AI one-to-one preprocessor 2112 , a first encoder 2114 , a data processor 2116 , and an AI setting unit 2118 . 21 shows the AI encoder 2110 and the transmitter 2130 as separate devices, the AI encoder 2110 and the transmitter 2130 may be implemented through one processor. In this case, it may be implemented as a dedicated processor, or it may be implemented through a combination of an AP or a general-purpose processor such as a CPU or GPU and S/W. In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

Also, the AI encoder 2110 and the transmitter 2130 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU. In one embodiment, the first encoder 2114 is configured as a first processor, and the AI one-to-one preprocessor 2112 , the data processor 2116 , and the AI setting unit 2118 include a second processor that is different from the first processor. It is implemented as a processor, and the transmitter 2130 may be implemented as a third processor different from the first processor and the second processor.

The AI encoding unit 2110 performs AI one-to-one preprocessing 2010 of the original image 2005 and first encoding of the first image 2015, and transmits the AI encoded data to the transmission unit 2130 . The transmitter 2130 transmits the AI encoded data to the AI decoding apparatus 200 .

The image data includes data obtained as a result of the first encoding of the first image 2015 . The image data may include data obtained based on pixel values in the first image 2015 , for example, residual data that is a difference between the first image 2015 and prediction data of the first image 2015 . . Also, the image data includes information used in the first encoding process of the first image 2015 . For example, the image data may include prediction mode information used to first encode the first image 2015, motion information, and quantization parameter related information used to first encode the first image 2015. .

The AI data includes information enabling the AI upscaling unit 236 to AI upscale the second image 2035 to an upscaling target corresponding to the one-to-one preprocessing target of the third DNN. In an example, the AI data may include resolution information of the first image 2015 that has been pre-processed by AI one-to-one. Also, the AI data may include information related to the first image 2035 . The information related to the first image 2035 is used for the resolution of the first image 2035 , the bit rate of image data obtained as a result of the first encoding of the first image 2035 , and the first encoding of the first image 2035 . information on at least one of the specified codec types may be included.

In one embodiment, the AI data includes an identifier of mutually agreed DNN setting information so that the second image 2035 can be AI upscaled to an upscale target corresponding to the one-to-one preprocessing target of the third DNN. can

Also, in an embodiment, the AI data may include DNN configuration information settable in the second DNN.

The AI 1-to-1 preprocessor 2112 may acquire the AI 1-to-1 preprocessed first image 2015 from the original image 2005 through the third DNN. The AI one-to-one preprocessor 2112 may AI downscale the original image 2005 by using the DNN setting information provided from the AI setting unit 2118 . The AI setting unit 2118 may determine a one-to-one preprocessing target of the original image 2005 based on a predetermined criterion.

In order to obtain the first image 2015 matching the one-to-one pre-processing target, the AI setting unit 2118 may store a plurality of settable DNN setting information in the third DNN. The AI setting unit 2118 obtains DNN setting information corresponding to a one-to-one pre-processing target among a plurality of DNN setting information, and provides the obtained DNN setting information to the AI one-to-one pre-processing unit 2112 .

Each of the plurality of DNN setting information may be trained to acquire a first image 2015 in which detailed features of the original image are well preserved and/or a first image 2015 having a predetermined image quality. For example, the DNN setting information of any one of the plurality of DNN setting information is the first image 2015 in which the detailed characteristics of the image are well preserved while maintaining the resolution of the original image 2005, for example, 2K ( Although the resolution of the original image 2005 of 2048*1080 is the same, information for acquiring the first image 2015 in which detailed characteristics of the original image 2005 are well preserved may be included. Here, the detailed feature of the original image may be a part with high spatiotemporal complexity. The other DNN setting information is the same as the resolution of the first image 2015, for example, the original image 2005 of 2K (2048*1080), but with the same resolution but improved image quality while maintaining the resolution of the original image 2005 as it is. Information for acquiring the improved first image 2015 may be included.

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 2118 may provide the DNN setting information obtained by combining some selected among the lookup table values to the AI one-to-one preprocessing unit 2112 according to the one-to-one preprocessing target.

Depending on the implementation, the AI setting unit 2118 may determine the structure of the DNN corresponding to the one-to-one preprocessing target, and obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel. have.

The plurality of DNN setting information for AI one-to-one preprocessing of the original image (2005) is obtained by using the DNN setting information of the second DNN obtained after the first DNN and the second DNN are jointly trained to form the third DNN and the second DNN. 2 As the DNN is jointly trained, it can have an optimized value. Here, each DNN configuration information includes at least one of the number of convolution layers included in the third DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel.

The AI one-to-one preprocessor 2112 sets the third DNN as the DNN setting information determined for the AI one-to-one preprocessing of the original image 2005, and the first image in which detailed features of the original image 2005 are preserved. (2015) and/or the first image 2015 of a predetermined quality may be obtained through the third DNN. When DNN setting information for AI one-to-one pre-processing of the original image 2005 is obtained among a plurality of DNN setting information, each layer in the third DNN processes the input data based on the information included in the DNN setting information. can

Hereinafter, a method for the AI setting unit 2118 to determine a one-to-one pre-processing target will be described. The one-to-one pre-processing target determines, for example, whether to acquire the first image 2015 that preserves the detailed characteristics of the original image to some extent while maintaining the resolution of the original image 2005 or the resolution of the original image 2005 . It may indicate how much the first image 2015 with improved image quality should be acquired while maintaining it.

The AI setting unit 2118 acquires one or more pieces of input information. In an embodiment, the input information includes a target resolution of the first image 2015, a target bitrate of the image data, a bitrate type of the image data (eg, a variable bitrate type, a constant bitrate type, or an average bitrate type); Color format (luminance component, chrominance component, red component, green component, blue component, etc.) to which AI one-to-one preprocessing is applied, codec type for the first encoding of the first image 2015, compression history information, original image ( 2005) and the type of the original image 2005 may be included.

One or more pieces of input information may include information previously stored in the AI encoding apparatus 2100 or received from a user.

The AI setting unit 2118 controls the operation of the AI one-to-one preprocessing unit 2112 based on the input information. In one embodiment, the AI setting unit 2118 determines a one-to-one pre-processing target according to the input information, and provides DNN setting information corresponding to the determined one-to-one pre-processing target to the AI one-to-one pre-processing unit 2112 . can

In an embodiment, the AI setting unit 2118 transmits at least a portion of the input information to the first encoder 2114 so that the first encoder 2114 sets the bitrate of a specific value, the bitrate of a specific type, and the specific codec. may cause the first image 2015 to be first encoded.

In one embodiment, the AI setting unit 2118 may be configured to at least one of a compression rate (eg, target bitrate), compression quality (eg, bitrate type), compression history information, and the type of the original image 2005 . Based on the one-to-one pre-processing target can be determined.

In one example, the AI setting unit 2118 may determine a one-to-one preprocessing target based on a preset or a compression rate or compression quality input from a user.

As another example, the AI setting unit 2118 may determine a one-to-one preprocessing target using compression history information stored in the AI encoding apparatus 2100 . For example, according to the compression history information available to the AI encoding apparatus 2100, the encoding quality or compression ratio preferred by the user may be determined, and the one-to-one preprocessing target may be determined based on the encoding quality determined based on the compression history information. This can be determined. For example, the resolution and quality of the first image 2015 may be determined according to the encoding quality that has been most frequently used according to the compression history information.

As another example, the AI setting unit 2118 determines the encoding quality that has been used more than a predetermined threshold value (eg, average quality of encoding quality that has been used more than a predetermined threshold value) according to the compression history information. Based on the one-to-one pre-processing target may be determined.

As another example, the AI setting unit 2118 may determine a one-to-one preprocessing target based on the resolution and type (eg, file format) of the original image 2005 .

In an embodiment, when the original image 2005 consists of a plurality of frames, the AI setting unit 2118 independently acquires DNN setting information for each predetermined number of frames, and sets the independently acquired DNN setting information to one AI. 1 may be provided to the pre-processing unit 2112 .

In one example, the AI setting unit 2118 may classify frames constituting the original image 2005 into a predetermined number of groups, and independently obtain DNN setting information for each group. The same or different DNN configuration information may be obtained for each group. The number of frames included in the groups may be the same or different for each group.

In another example, the AI setting unit 2118 may independently determine DNN setting information for each frame constituting the original image 2005 . The same or different DNN configuration information may be obtained for each frame.

An exemplary structure of the third DNN 1610, which is the basis of the AI one-to-one preprocessing, has been described above with reference to FIG. 14, and thus will be omitted.

Referring back to FIG. 21 , the AI setting unit 2118 transmits AI data to the data processing unit 2116 . The AI data includes information enabling the AI upscaling unit 236 to AI upscale the second image 2035 to an upscaling target corresponding to the one-to-one preprocessing target of the third DNN. The first encoder 2114 receiving the one-to-one pre-processed first image 2015 from the AI one-to-one preprocessor 2112 converts the first image 2015 into the first image according to the frequency transformation-based image compression method. By encoding, the amount of information in the first image 2015 can be reduced. As a result of the first encoding through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1), image data is obtained. The image data is obtained according to a rule of a predetermined codec, that is, a syntax. For example, the image data includes residual data that is a difference between the first image 2015 and the prediction data of the first image 2015, prediction mode information used to first encode the first image 2015, motion information, and Information related to a quantization parameter used for first encoding the first image 2015 may be included. The image data obtained as a result of the first encoding by the first encoder 2114 is provided to the data processor 2116 .

The data processing unit 2116 generates AI encoded data including the image data received from the first encoding unit 2114 and the AI data received from the AI setting unit 2118 .

In an embodiment, the data processing unit 2116 may generate AI-encoded data including image data and AI data in a separate state. As an example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream.

In another embodiment, the data processing unit 2116 may include AI data in image data obtained as a result of the first encoding by the first encoding unit 2114 and generate AI encoded data including the corresponding image data. . For example, the data processing unit 2116 may generate image data in the form of one bitstream by combining a bitstream corresponding to image data and a bitstream corresponding to AI data. To this end, the data processing unit 2116 may express AI data as bits having a value of 0 or 1, that is, a bitstream. In an embodiment, the data processing unit 2116 may include a bitstream corresponding to AI data in supplemental enhancement information (SEI) that is an additional information area of a bitstream obtained as a result of the first encoding.

AI-encoded data is transmitted to the transmitter 2130 . The transmitter 2130 transmits the AI encoded data obtained as a result of the AI encoding through a network.

In one embodiment, the AI encoded data is a hard disk, a magnetic medium such as a floppy disk and magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk. medium) may be stored in a data storage medium including

After the first DNN 800 and the second DNN 300 are jointly trained, the method of jointly training the third DNN 1610 with the second DNN 300 is omitted as described above in FIGS. 16 and 17 . .

Meanwhile, the above-described embodiments of the present disclosure can be written as a program or instruction that can be executed in a computer, and the written program or instruction can be stored in a medium.

The medium may continuously store a program or instructions executable by a computer, or may be a temporary storage for execution or download. In addition, the medium may be various recording means or storage means in the form of a single or several hardware combined, it is not limited to a medium directly connected to any computer system, and may exist distributed on a network. Examples of the medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk, and those configured to store program instructions, including ROM, RAM, flash memory, and the like. In addition, examples of other media may include recording media or storage media managed by an app store that distributes applications, sites that supply or distribute various other software, or servers.

Meanwhile, the above-described DNN-related model may be implemented as a software module. When implemented as a software module (eg, a program module including instructions), the DNN model may be stored in a computer-readable recording medium.

In addition, the DNN model may be integrated in the form of a hardware chip to become a part of the aforementioned AI decoding apparatus 200 or AI encoding apparatus 600 . For example, a DNN model may be built in the form of a dedicated hardware chip for artificial intelligence, or as part of an existing general-purpose processor (eg, CPU or application processor) or graphics-only processor (eg, GPU). it might be

In addition, the DNN model may be provided in the form of downloadable software. The computer program product may include a product (eg, a downloadable application) in the form of a software program distributed electronically through a manufacturer or an electronic market. For electronic distribution, at least a portion of the software program may be stored in a storage medium or may be temporarily generated. In this case, the storage medium may be a server of a manufacturer or an electronic market, or a storage medium of a relay server.

In the above, the technical idea of the present disclosure has been described in detail with reference to preferred embodiments, but the technical idea of the present disclosure is not limited to the above embodiments, and those of ordinary skill in the art within the scope of the technical spirit of the present disclosure Various modifications and changes are possible by the person.

Claims (15)

a memory storing one or more instructions; and
a processor executing the one or more instructions stored in the memory;
The processor is
Determining the size of the resolution of the original video,
If the resolution of the original image is greater than a predetermined value, the AI downscaled first image is acquired from the original image through DNN for downscaling,
When the resolution of the original image is less than or equal to the predetermined value, the AI 1-to-1 pre-processed first image is obtained from the original image through DNN for one-to-one pre-processing for upscaling,
generating image data by first encoding the first image;
Transmitting AI data and the image data including information related to the AI downscaling or information related to the AI one-to-one preprocessing,
The AI data includes information for selecting DNN setting information of the DNN for upscaling for AI upscaling of the second image, and the second image is generated through the first decoding of the image data,
The DNN setting information of the DNN for downscaling is obtained through a first joint training of the DNN for downscaling and the DNN for upscaling for AI upscaling of the second image,
The DNN setting information of the DNN for the one-to-one preprocessing is the first of the DNN for the one-to-one preprocessing and the DNN for the upscaling by using the DNN setting information of the DNN for the upscaling obtained through the first joint training. 2 Acquired through joint training,
The AI encoding apparatus of an image, characterized in that if the parameters of the DNN for upscaling are updated in the first joint training process, the parameters of the DNN for one-to-one preprocessing are also updated in the second joint training process. According to claim 1,
The AI encoding apparatus of an image, wherein the DNN for one-to-one preprocessing enhances features of the original image while maintaining the same resolution as that of the original image. According to claim 1,
The DNN for the downscaling and the DNN for the upscaling are,
AI encoding apparatus, characterized in that trained based on the first loss information corresponding to the comparison result between the first training image output from the upscale DNN and the original training image before AI downscale. 4. The method of claim 3,
The downscale DNN is,
Second loss information corresponding to a comparison result between the reduced training image corresponding to the original training image and the second training image output from the downscale DNN and third loss information corresponding to the spatial complexity of the second training image AI encoding apparatus, characterized in that it is trained by further considering at least one of. According to claim 1,
The DNN for the one-to-one preprocessing is,
Taking the reduced training image corresponding to the original training image as input, the fourth loss information corresponding to the comparison result between the fourth training image and the original training image output from the one-to-one preprocessing DNN and the upscaling DNN AI encoding device, characterized in that it is trained in consideration. According to claim 1,
The AI encoding device, characterized in that the resolution of the predetermined value is 2K (2048 * 1080). According to claim 1,
The AI encoding apparatus, characterized in that the resolution of the predetermined value is 4K (4096 * 2160). According to claim 1,
In the first joint training process, when parameters of any one of the DNN for upscaling and the DNN for downscaling are updated, parameters of the other DNN are also updated. delete a memory storing one or more instructions; and
a processor executing the one or more instructions stored in the memory;
The processor is
Obtaining AI data related to AI downscaling or AI one-to-one preprocessing from the image data generated as a result of the first encoding of the first image and the original image to the first image,
obtaining a second image corresponding to the first image by first decoding the image data;
Obtaining DNN setting information for AI upscaling of the second image among a plurality of DNN setting information, based on the AI data,
An AI upscaled third image is generated from the second image through a deep neural network (DNN) for upscaling that operates with the obtained DNN setting information,
The plurality of DNN setting information includes a first joint training of the DNN for upscaling and the DNN for downscaling used for AI downscaling of the original video,
Using the DNN setting information for AI upscaling obtained as a result of the first linked training, the second linked training of the DNN for one-to-one preprocessing used for AI one-to-one preprocessing of the original image and the DNN for upscaling obtained through
The AI decoding apparatus according to claim 1, wherein when the parameters of the upscaling DNN are updated in the first joint training process, the parameters of the one-to-one preprocessing DNN are also updated in the second joint training process. 11. The method of claim 10,
The AI decoding apparatus of an image, wherein the DNN for one-to-one preprocessing enhances features of the original image while maintaining the same resolution as that of the original image. In the AI encoding method of an image,
determining the resolution of the original image;
if the resolution of the original image is less than or equal to a predetermined value, acquiring a first image preprocessed by AI one-to-one through DNN for one-to-one preprocessing for upscaling;
if the resolution of the original image is greater than the predetermined value, obtaining an AI downscaled first image from the original image through a downscaling DNN;
generating image data by first encoding the first image; and
Transmitting AI data and the image data including information related to the AI one-to-one pre-processing or information related to the AI downscaling,
The AI data includes information for selecting DNN setting information of the DNN for upscaling for AI upscaling of the second image, and the second image is generated through the first decoding of the image data,
The DNN setting information of the DNN for downscaling is obtained through a first joint training of the DNN for downscaling and the DNN for upscaling for AI upscaling of the second image,
The DNN setting information of the DNN for the one-to-one preprocessing is the first of the DNN for the one-to-one preprocessing and the DNN for the upscaling by using the DNN setting information of the DNN for the upscaling obtained through the first joint training. 2 Acquired through joint training,
The AI encoding method of an image, characterized in that if the parameters of the DNN for upscaling are updated in the first joint training process, the parameters of the DNN for one-to-one preprocessing are also updated in the second joint training process. 13. The method of claim 12,
The AI encoding method of an image, wherein the DNN for one-to-one preprocessing enhances features of the original image while maintaining the same resolution as that of the original image. In the AI decoding method of an image,
acquiring AI data related to AI downscaling or AI one-to-one preprocessing from the image data generated as a result of the first encoding of the first image and the original image to the first image;
obtaining a second image corresponding to the first image by first decoding the image data;
obtaining DNN setting information for AI upscaling of the second image from among a plurality of DNN setting information, based on the AI data; and
Generating an AI upscaled third image from the second image through a deep neural network (DNN) for upscaling that operates with the obtained DNN setting information,
The plurality of DNN setting information includes a first associated training of the DNN for upscaling and the DNN for downscaling used for AI downscaling of the original image,
Using the DNN setting information for AI upscaling obtained as a result of the first linked training, the second linked training of the DNN for one-to-one preprocessing used for AI one-to-one preprocessing of the original image and the DNN for upscaling obtained through
In the first joint training process, if the parameters of the DNN for upscaling are updated, the parameters of the DNN for one-to-one preprocessing are also updated in the second joint training process. A computer-readable recording medium recording a program for executing the method of claim 12.

Images