WO2019209008A1 - System for improving video quality by using changed macroblock extraction technique - Google Patents
System for improving video quality by using changed macroblock extraction technique Download PDFInfo
- Publication number
- WO2019209008A1 WO2019209008A1 PCT/KR2019/004895 KR2019004895W WO2019209008A1 WO 2019209008 A1 WO2019209008 A1 WO 2019209008A1 KR 2019004895 W KR2019004895 W KR 2019004895W WO 2019209008 A1 WO2019209008 A1 WO 2019209008A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- learning
- video
- file
- downscaling
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/21—Server components or server architectures
- H04N21/222—Secondary servers, e.g. proxy server, cable television Head-end
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/47—End-user applications
- H04N21/472—End-user interface for requesting content, additional data or services; End-user interface for interacting with content, e.g. for content reservation or setting reminders, for requesting event notification, for manipulating displayed content
Definitions
- the present invention relates to a video quality improvement system using a change macroblock extraction technique.
- the Fourth Industrial Revolution is triggering a new era of broadcasting services and is changing the media landscape.
- the Internet broadcasting platform anyone has the environment to broadcast, and as the number of Internet broadcasting viewers is increasing, new media platforms such as Netflix, YouTube, Hulu, and Amazon TV are booming.
- the present invention is designed to convert one video data into the form of a downscaling file having reduced resolution and capacity, and to restore the downscaling file to the original level through an artificial neural network-based operation. .
- the video quality improvement system is a video service server for transmitting original video data to an AI image processing server to perform AI video learning of video data, and AI video to the original video data received from the video service server.
- AI image learning is performed by separately classifying an image area and an image area where a change is not detected, and transmitting the generated downscaling file and the neural network file to the video service server, and in response to the image restoration request, the neural network file.
- Phase through AI operation It may include an AI image processing server for restoring the resolution of the downscaling file.
- the capacity and resolution of the video data are reduced and stored in the form of a downscaling file, more video data can be stored in a storage device having the same storage space. This reduces the cost of expanding storage space.
- the present invention restores the resolution at the time of actually playing the video, it is possible to obtain the same video quality as the original data with a relatively small capacity.
- the present invention can reduce the time and data usage required for image learning and reconstruction, since only the image region of the changed screen is image learning and image reconstruction separately.
- FIG. 1 is a diagram illustrating a configuration of a video quality improvement system according to an exemplary embodiment of the present invention.
- FIG. 2 is a diagram illustrating a configuration of an AI image processing server according to an exemplary embodiment of the present invention.
- FIG. 3 is a diagram illustrating a configuration of a codec support unit according to an exemplary embodiment of the present invention.
- FIG. 4 is a diagram illustrating a configuration of an AI learning support unit according to an exemplary embodiment of the present invention.
- FIG. 5 illustrates a video frame image displaying a motion vector according to an embodiment of the present invention.
- FIG. 6 is a flowchart illustrating a procedure of performing a restoration operation in a video quality improvement system according to an exemplary embodiment of the present invention.
- FIG. 7 is a flowchart illustrating a process of an image learning operation according to an exemplary embodiment of the present invention.
- the video quality improvement system is a video service server for transmitting original video data to an AI image processing server to perform AI video learning of video data, and AI video to the original video data received from the video service server.
- AI image learning is performed by separately classifying an image area and an image area where a change is not detected, and transmitting the generated downscaling file and the neural network file to the video service server, and in response to the image restoration request, the neural network file.
- Phase through AI operation It may include an AI image processing server for restoring the resolution of the downscaling file.
- FIG. 1 is a diagram illustrating a configuration of a video quality improvement system according to an exemplary embodiment of the present invention.
- the video quality improvement system 10 may include an AI image processing server 100, a video service server 200, and a user device 300.
- the video service server 200 may store video data uploaded from a plurality of users or administrators, and may be a platform server that provides a service for transmitting specific video data to a user in response to a user request.
- the video service server 200 constituting the system 10 of the present invention may mean not only a video service company server but also a server functioning to store video content such as a cloud server of a user.
- the video service server 200 may be in an interlocked state with the image processing server 100 according to an exemplary embodiment of the present invention. Accordingly, the video service server 200 transmits specific video data (original video data before video learning) to the AI image processing server 100 at a time when a predetermined condition is met, and may request video learning.
- the video service server 200 may be a server 200 that implements a platform of obtaining video content from a plurality of users and providing the obtained video data to a plurality of other users.
- the video service server 200 may immediately request video learning to the AI image processing server 100 as a specific video is uploaded from an arbitrary user.
- the AI image processing server 100 may perform image learning on a video obtained from the video service server 200.
- the image learning refers to an operation of generating a neural network file required for restoring a low quality file having a resolution lower than a predetermined reference value to an original image level (for example, a level that matches the reference image with a preset reference value or more). can do.
- the AI image processing server 100 may generate a downscaling file having a reduced resolution and a neural network file including metadata used to restore an image as a result of image learning of an original video (high resolution data of a reference value or more). Can be.
- the neural network file may be applied to the downscaling file later to include meta information necessary to perform AI-based image reconstruction.
- the neural network file may include information related to various parameters (weight, bias, etc.) of the artificial neural network required when restoring the downscaling file to the original file.
- the AI image processing server 100 may store the downscaling file calculated as the image learning result in the video service server 200 and the neural network file corresponding thereto in the AI image processing server 100.
- the AI image processing server 100 receives only the downscaling file from the video service server 200 later, and stores the neural network file corresponding to the received downscaling file. After extraction, image restoration may be performed.
- the video service server 200 may transmit a downscaling file to the user device 300, and the AI image processing server 100 may deliver a neural network file to the user device 300. .
- the AI image processing server 100 may provide both the downscaling file and the neural network file calculated as a result of image learning to the video service server 200, and the video service server 200 may store both types of files.
- the video service server 200 may immediately provide a downscaling file and a neural network file for the corresponding content to the AI image processing server 100 and request to restore the image quality.
- the video content restored in the AI image processing server 100 may be provided to the user device 300 through the video service server 200 or may be directly provided to the user device 300.
- the user device 300 may be in a state in which a program that supports access to the AI image processing server 100 and data transmission / reception is installed at the request of the video service server 200.
- the video service server 200 may transmit both the downscaling file and the neural network file to the user device 300.
- the video service server 200 may determine whether to transmit the downscaling file and the neural network file to the user device 300 and the format of the transmission data based on specifications of the GPU of the user device 300.
- the method of storing the image learning result file (downscaling file and neural network file) and providing the device between the devices is not limited to the above-described form, and various methods may be applied.
- the method of storing the image learning result file and providing the device-to-device may vary according to the state of the user device 300, the model, the interworking format of the video service server 200, and the AI image processing server 100.
- FIG. 2 is a diagram illustrating a configuration of an AI image processing server according to an exemplary embodiment of the present invention.
- the AI image processing server 100 will be referred to as a processing server 100.
- the processing server 100 may include a communication unit 110, a storage unit 120, and a controller 130.
- the communication unit 110 may use a network for data transmission and reception between a user device and a server, and the type of the network is not particularly limited.
- the network may be, for example, an IP (Internet Protocol) network providing a transmission / reception service of a large amount of data through an Internet protocol (IP), or an All IP network integrating different IP networks.
- IP Internet Protocol
- the network may include a wired network, a wireless broadband network, a mobile communication network including WCDMA, a high speed downlink packet access (HSDPA) network, and a long term evolution (LTE) network, LTE advanced (LTE-A). ), Or one of a mobile communication network including 5G (Five Generation), a satellite communication network, and a Wi-Fi network, or a combination of at least one of them.
- 5G Wireless Generation
- satellite communication network and a Wi-Fi network
- the communication unit 110 may perform data communication with the video service server 200 and the user device 300.
- the storage unit 120 may include, for example, an internal memory or an external memory.
- the internal memory may be, for example, volatile memory (for example, dynamic RAM (DRAM), static RAM (SRAM), or synchronous dynamic RAM (SDRAM), etc.), non-volatile memory (for example, OTPROM (one). time programmable ROM (PROM), programmable ROM (PROM), erasable and programmable ROM (EPROM), electrically erasable and programmable ROM (EEPROM), mask ROM, flash ROM, flash memory (such as NAND flash or NOR flash), hard drives, Or it may include at least one of a solid state drive (SSD).
- volatile memory for example, dynamic RAM (DRAM), static RAM (SRAM), or synchronous dynamic RAM (SDRAM), etc.
- non-volatile memory for example, OTPROM (one).
- the external memory may be a flash drive such as compact flash (CF), secure digital (SD), micro secure digital (Micro-SD), mini secure digital (mini-SD), extreme digital (XD), It may further include a multi-media card (MMC) or a memory stick.
- the external memory may be functionally and / or physically connected to the electronic device through various interfaces.
- the storage unit 120 may store a downscaling file that is a result of performing image learning and a neural network file that is a file required to restore the downscaling file to an original level later.
- the storage unit 120 may store an algorithm required for the downscaling file, an image learning operation for generating a neural network file, and an artificial neural network operation for reconstructing a resolution from the downscaling file.
- the controller 130 may also be referred to as a processor, a controller, a microcontroller, a microprocessor, a microcomputer, or the like.
- the control unit may be implemented by hardware (hardware) or firmware (firmware), software, or a combination thereof.
- an embodiment of the present invention may be implemented in the form of a module, procedure, function, etc. that performs the functions or operations described above.
- the software code may be stored in a memory and driven by the controller.
- the memory may be located inside or outside the user terminal and the server, and may exchange data with the controller by various known means.
- the controller 130 may include a codec support unit 131, an AI learning support unit 132, and an AI image restoration unit 133.
- the codec support unit 131 may provide a general function of a codec for compressing or decompressing a video.
- the codec support unit 131 may perform data extraction and keyframe extraction of a video acquired for image learning using a codec function.
- the extracted keyframes may be processed separately from the remaining frames, which are frames other than the keyframes, in image learning.
- the AI learning support unit 132 may support AI based image learning.
- the AI learning support unit 132 may generate a downscaling file from the obtained video file and a neural network file, which is a file required to restore the downscaling file to the original level.
- the AI learning support unit 132 may perform image learning by dividing the key frame extracted by the codec support unit 131 and the remaining frames except the key frame.
- the AI learning support unit 132 may perform image learning on the entire image belonging to the key frame when the image learning is performed on the key frame.
- the AI learning support unit 132 When the AI learning support unit 132 performs the image learning on the remaining frames which are not extracted as key frames, the AI learning support unit 132 excludes the macro blocks having the same value as the key frames and targets the macro blocks having different values from the key frames. You can do some learning.
- the AI image reconstructor 133 may upscale the downscaling file generated as a result of image learning based on the neural network file.
- the AI image restoration unit 133 may provide the user device 300 or the video service server 200 with data whose upscaling is completed and whose quality is restored.
- the AI image restoration unit 133 may perform an image restoration operation by separating a key frame and a residual frame in the same manner as the image learning method.
- a key frame is first image-restored, and for a remaining frame, a macroblock having the same value as a keyframe (hereinafter, referred to as a still macroblock) copies an image for a keyframe as it is, and has a value different from that of the keyframe.
- the macroblock (hereinafter, referred to as a change macroblock) may perform an additional image reconstruction operation. Accordingly, the data load and the calculation speed required for the image restoration operation can be reduced.
- the AI image reconstructor 133 may determine whether a match rate with the original data is greater than or equal to a reference value, and provide feedback information to the AI learning support unit 132. According to various embodiments of the present disclosure, the AI image reconstruction unit 133 may transmit the information about the matching rate to the video service server 100 and, when the matching rate is lower than the reference value, may request to provide data for further image learning.
- the AI image reconstructor 133 may reconstruct an image by dividing a key frame and a remaining frame of the video.
- image restoration is performed for the entire image region.
- image restoration is performed only for some image regions (eg, change image object regions not overlapping with key frames). Can be pasted from the reconstructed image of the key frame. This re-uses the previously reconstructed image when reconstructing an image of the remaining frame having an image region overlapping with the key frame, thereby preventing unnecessary reconstruction operation and reducing the time required for the reconstruction process.
- FIG. 3 is a diagram illustrating a configuration of a codec support unit according to an exemplary embodiment of the present invention.
- the codec supporter 131 may include a data determiner 131a, a frame extractor 131b, and a macroblock determiner 131c.
- the data determiner 131a may extract detailed information related to image learning or image reconstruction for the received video based on the function of the codec.
- the detailed information may include, for example, information related to basic video playback such as a basic resolution, a compression method, and a playback time of the video file received from the video service server 200.
- the frame extractor 131b may extract a keyframe from a video based on the function of the codec. In other words, the frame extractor 131b may distinguish between a key frame and a remaining frame other than the key frame among all the frames constituting the video. In this case, the key frame may be separated from the remaining frame to perform a separate image learning process.
- the frame extractor 131b may provide the extracted key frame information to the AI learning support unit 132 or the AI image restoration unit 133 for immediate image learning or image restoration.
- the frame extractor 131b may select and extract, as a keyframe, the first frame in which the main scene change has started, based on the basic functions provided by the codec.
- the macroblock determination unit 131c may perform an operation of dividing a macroblock into a change macroblock and a still macroblock with respect to the remaining frames except for the keyframe.
- the division information on the change macroblock and the still macroblock may be calculated by the macroblock determination unit 131c and then provided to the AI learning support unit or the AI image reconstruction unit.
- the macroblock may mean a unit in which several pixel blocks are grouped for motion compensation and motion prediction in main image compression schemes such as H.261, H.263, and MPEG, and may include luminance signals and color difference signals.
- the existing codec (eg, H.264) is a lossy extrusion by removing the overlapped information in the image (motion compensation, spatial (dct, quantization)).
- motion compensation spatial (dct, quantization)
- the previous frame stores the entire image in a file, but the current frame after the current frame is compared to the previous frame. This method reduces the file size by saving only the vector and the residual image as a file.
- the building which is the background of the person, will move without moving, and only the face of the person will move, so that the face area of the person will have a predetermined motion vector value.
- the area will have a value close to zero.
- the codec may perform an operation (quantization) of replacing a region having a small value close to zero as a region overlapping with a previous frame and quantizing the region, thereby performing lossy compression.
- the macroblock determination unit 131c may calculate macroblocks in which a motion vector is considered to be 0 and macroblocks having a significant value by using a compression principle of the basic codec. Can be.
- a macroblock in which a motion vector is considered to be 0 is a still macroblock
- a macroblock having a motion vector having a significant value means a change macroblock.
- the codec support unit 131 may perform various operations based on a conventional codec function, in addition to a keyframe extraction operation and a classification operation of a change macroblock and a still macroblock in a remaining frame.
- the codec support unit 131 may determine a complexity of an image determined from a key frame and distinguish a genre of a corresponding video in relation to an operation of extracting a key frame from a codec. For example, the codec support unit 131 may distinguish between an animation genre and a live action genre.
- the genre classification information calculated according to the genre classification operation may be provided to the AI learning support unit and the AI image reconstruction unit 133 for effective AI learning.
- FIG. 4 is a diagram illustrating a configuration of an AI learning support unit according to an exemplary embodiment of the present invention.
- the AI learning support unit 132 may include a change object extractor 132a, a main object extractor 132b, and an AI learning performer 132c.
- the change object extractor 132a may perform an operation of determining the change object based on the change macroblock information of the remaining frame calculated by the codec support unit 131. To describe the change object determination operation, reference is made to FIG. 5.
- FIG. 5 illustrates a video frame image displaying a motion vector according to an embodiment of the present invention.
- a video frame image has an image region 510 composed of a macroblock in which a motion vector (arrow) exists and a motion vector does not exist or is less than a reference value. It may be divided into an image area 520.
- the change object extracting unit 132a may determine the change image object composed of the change macroblock and the change macroblock based on the presence or absence of the motion vector value.
- the change object extractor 132a may determine a still image object including the still macroblock and the still macroblock based on the presence or absence of a motion vector value or the magnitude of the value.
- the main object extractor 132b may perform an operation of extracting the image object based on the importance determination instead of moving the image object.
- the main object extractor 132b may extract the object separately, even if it is determined that the object has importance in the corresponding video data even in the still image.
- the extracted image object can be separately performed AI image learning later.
- the main object extractor 132b may include a size of an image object identified in a key frame image, an image complexity of the image object (based on the variety of colors and the complexity of an outline), a person (increasing importance in the case of a person), The main object can be determined based on various criteria such as focusing (increasing importance when the focus exists). For example, the main object extractor 132b may have a ratio of the area occupied by the image of the book or the size of the image object of the book even if the image of a specific object (eg, a book) displayed at the center of the frame is still. If it is more than the reference value, it can be determined that there is importance and can be extracted as a main object accordingly.
- a specific object eg, a book
- the main object extractor 132b may set the importance value to a higher value as the size value of the image object increases. If the importance of the calculated image object is greater than or equal to the reference value, the corresponding image object may be selected as the main image object.
- the AI learning performer 132c may support a compression function through AI-based image learning on the obtained video data. Specifically, the AI learning performer 132c may generate a downscaling file from the obtained video file and a neural network file that is a file required to restore the quality of the downscaling file to the original level later.
- the downscaling file refers to a compressed file
- the AI learning performer 132c may perform video file compression by performing such image learning.
- the video file downscaled and compressed by such AI image learning may be reconstructed (decompressed) by a reconstruction operation performed based on a neural network file in the AI image reconstruction unit 133.
- the AI learning performer 132c may perform separate image learning on the key frame and the remaining frame of the video.
- the AI learning performer 132c first performs image learning on the entire image area with respect to the key frame, and classifies at least two image areas with respect to the remaining frames according to an arbitrary criterion (duplicate or importance). Image learning of the image region is omitted. In the future, when reconstructing the resolution of the remaining frame, an image region in which image learning is not performed may be performed by pasting from the image reconstruction result of the key frame.
- the AI learning performer 132c may perform a separate AI image learning operation targeting the change image object extracted by the change object extractor 132a. Accordingly, the AI learning performer 132c may generate a separate neural network file for the change image object. Accordingly, when the AI image restoration unit 133 restores the image later, an image restoration operation for the change image object is performed based on a separate neural network file for the change image object, and the rest of the still image object (eg, a background) is performed. For example, the final reconstructed image may be generated by copying and pasting a duplicate image of a keyframe. The AI learning performer 132c performs a separate image learning method on the change image object, thereby reducing the time required and data capacity for the reconstructed image.
- the AI learning performing unit 132c performs a separate image learning on the main image object extracted by the main object extracting unit 132b, and as a result, the neural network required for later restoring the resolution of the main image object. You can create a file. Accordingly, during image restoration, an image restoration operation for the main image object may be generated based on a neural network file corresponding to the main image object.
- FIG. 6 is a flowchart illustrating a procedure of performing a restoration operation in a video quality improvement system according to an exemplary embodiment of the present invention.
- the video quality improvement system may be implemented through an AI image processing server 100, a video service server 200, and a user device 300.
- the video service server 200 may be implemented in the same device as the AI image processing server 100.
- the AI image processing server 100 and the video service server 200 will be exemplified as separate objects.
- the video service server 200 may mean, for example, an OTT (Over The Top) service server that provides image content using an internet communication network.
- the video service server 200 provides image data to the AI image processing server 100 to learn AI images of the owned or produced video and to generate a low-scaling downscaling file. Can be received.
- the video service server 200 may perform operation 605 of providing original image data to the AI image processing server 100. Thereafter, when the AI image processing server 100 receives the image data from the video service server 200, the AI image processing server 100 may perform operation 610 of performing AI image learning. Since the description of the AI image learning has been described above, it will be omitted.
- the AI image processing server 100 generates a result file (downscaling file and neural network file) as a result of image learning, and performs operation 615 of providing the same to the video service server 200.
- the video service server 200 may perform operation 620 of receiving a download request signal for the corresponding video from the user device 300 while having the downscaling file and the neural network file resulting from the image learning.
- the video service server 200 checks information such as the name of the video content requested to be downloaded from the user device 300, and sends corresponding learning result files (downscaling file and neural network file) to the AI image processing server 100 again.
- the operation 625 may be performed.
- the AI image processing server 100 may recognize this as an image restoration request signal, and may perform operation 630, which is an AI operation based image restoration operation. Thereafter, the AI image processing server 100 may perform operation 635 to deliver the restored file to the video service server 200, and then perform operation 640 to deliver the restored file to the user device 300 from the video service server 200. have.
- the video service server 200 intervenes so that the user device 300 directly accesses the AI image processing server 100, and the restored video file is directly transferred to the user device 300. Can be provided. Accordingly, when the user requests to play a specific video provided on the web or app provided by the video service server 200, the user may be immediately provided with the restored file from the AI image processing server 100 performing the restoration operation.
- a subject to perform an image restoration operation may be determined based on the specifications of the user device 300 according to various embodiments.
- the video service server 200 transmits the learning result file to the user device 300 when the user device 300 requests the user device 300 or determines that the user device 300 is suitable for AI calculation for image restoration.
- the image restoration operation may be performed at.
- the video service server 200 may deliver a downscaling file and a neural network file, which are image learning results, instead of original video data or restored data.
- the user may provide the learning result file to the AI image processing server 100 by, for example, playing it through a player provided by the AI image processing server 100 when the user device 300 actually wants to play the video data. have.
- the AI image processing server 100 may perform an image restoration operation through the learning result file received from the user device 300.
- FIG. 7 is a flowchart illustrating a process of an image learning operation according to an exemplary embodiment of the present invention.
- the controller 130 of the image processing server 100 may acquire image data from the outside and perform operation 705 to confirm that image data is obtained. Thereafter, the controller 130 may perform operation 710 of extracting a keyframe of video data based on the codec.
- the image processing server 100 may include various codec information for decoding video data having various extensions, and may extract a key frame of a video obtained based on the codec.
- controller 130 may perform operation 715 for performing image learning on the extracted key frame.
- the controller 130 may perform image learning on a remaining frame that is a frame other than a key frame, but may perform separate image learning on an image area different from the key frame except for a portion overlapping with the key frame. To this end, the controller 130 may perform operation 720 to extract the change macroblock of the remaining frame.
- the change macroblock is an object extracted to determine and separate a region where a motion occurs in the next scene when compared with a key frame. Thereafter, the controller 130 may perform operation 725 for extracting the change macroblock separately and performing image learning thereof.
- the order of operations 715 and 720 may be changed.
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Description
Claims (7)
- 동영상 데이터의 AI 영상 학습을 수행하기 위해 AI 영상 처리 서버로 원본 동영상 데이터를 전송하는 동영상 서비스 서버;A video service server for transmitting original video data to an AI video processing server to perform AI video learning of video data;상기 동영상 서비스 서버로부터 수신한 원본 동영상 데이터를 AI 영상 처리하여 상기 원본 동영상 데이터에 대응하는 다운스케일링 파일을 생성하고, 상기 다운스케일링 파일의 해상도를 복원하는 데 요구되는 파일인 신경망 파일을 생성하는 AI 영상 학습 동작을 수행하되, 동영상 데이터에서 변화가 감지되는 이미지 영역과 변화가 감지되지 않는 이미지 영역을 각각 별도로 구분하여 AI 영상 학습을 수행하고, 생성된 상기 다운스케일링 파일 및 상기 신경망 파일을 상기 동영상 서비스 서버로 전송하며, AI image processing of original video data received from the video service server to generate a downscaling file corresponding to the original video data, and to generate a neural network file which is a file required to restore the resolution of the downscaling file. A learning operation is performed, and the AI image learning is performed by separately distinguishing the image region where the change is detected from the image data and the image region where the change is not detected, and generating the downscaling file and the neural network file from the video service server. To,영상 복원 요청에 대응하여 상기 신경망 파일을 이용한 AI 연산을 통해 상기 다운스케일링 파일의 해상도를 복원하는 AI 영상 처리 서버;를 포함하는 것을 특징으로 하는 동영상 화질 개선 시스템.And an AI image processing server reconstructing the resolution of the downscaling file through an AI operation using the neural network file in response to an image restoration request.
- 제 1항에 있어서,The method of claim 1,상기 동영상 서비스 서버에 동영상 다운로드 요청 신호를 전달하는 사용자 기기;를 더 포함하고, And a user device for transmitting a video download request signal to the video service server.상기 동영상 서비스 서버는 The video service server상기 사용자 기기로부터 동영상 데이터의 다운로드 요청 신호를 수신하면 영상 복원을 위해, 다운로드 요청된 동영상 데이터에 대응하는 다운스케일링 파일 및 신경망 파일을 상기 AI 영상 처리 서버로 전송하는 것을 특징으로 하는 동영상 화질 개선 시스템. And a downscaling file and a neural network file corresponding to the downloaded video data are transmitted to the AI image processing server when the download request signal of the video data is received from the user device.
- 제 1항에 있어서, The method of claim 1,상기 AI 영상 처리 서버는 The AI image processing server상기 동영상 서비스 서버로부터 획득된 동영상 데이터에 대하여 코덱 기능을 수행하여, AI 영상 학습에 요구되는 데이터를 추출 및 분류하는 코덱 지원부;를 포함하고, And a codec support unit that performs a codec function on the video data obtained from the video service server, and extracts and classifies data required for AI video learning.상기 코덱 지원부는The codec support unit동영상 데이터의 키 프레임을 추출하여, 상기 키 프레임과 키 프레임을 제외한 나머지 프레임인 잔여 프레임을 분리하는 프레임 추출부;를 포함하는 것을 특징으로 하는 동영상 화질 개선 시스템. And a frame extractor which extracts a key frame of the video data and separates the key frame and the remaining frames other than the key frame.
- 제 3항에 있어서, The method of claim 3, wherein상기 코덱 지원부는The codec support unit상기 잔여 프레임 중 상기 키 프레임과 동일한 값을 갖는 매크로블록은 정지 매크로블록으로 판단하고, 상기 키 프레임과 상이한 값을 갖는 매크로블록은 변화 매크로블록으로 분류하는 매크로블록 판단부;를 포함하여 구성되는 것을 특징으로 하는 동영상 화질 개선 시스템.The macroblock having the same value as the key frame among the remaining frames is determined as a still macroblock, and the macroblock having a value different from the key frame is classified as a change macroblock. Featured video quality improvement system.
- 제 1항에 있어서,The method of claim 1,상기 AI 영상 처리 서버는The AI image processing serverAI 기반의 영상 학습을 수행하는 AI 학습 지원부;를 포함하되, Includes; AI learning support unit for performing AI-based image learning,상기 AI 학습 지원부는The AI learning support unit동영상 데이터의 키 프레임과 잔여 프레임에 대하여 각각 별도의 영상 학습을 수행하되, 키 프레임에 대하여는 전체 이미지 영역에 대한 영상학습을 수행하고, 잔여 프레임에 대하여는 이미지 영역을 적어도 2 종류의 영역으로 분류하고 그 중 일 이미지 영역에 대한 영상 학습은 생략하는 AI 학습 수행부;를 포함하는 것을 특징으로 하는 동영상 화질 개선 시스템.Perform separate image learning on the key frame and the remaining frame of the video data, perform image learning on the entire image area on the key frame, and classify the image area into at least two types of areas on the remaining frame. And an AI learning performing unit for skipping image learning of one image region.
- 제 5항에 있어서The method of claim 5상기 AI 학습 지원부는 The AI learning support unit상기 잔여 프레임의 이미지 영역에 대하여, 변화 매크로블록으로 구성되는 변화 이미지 객체 영역과 정지 매크로블록으로 구성되는 정지 이미지 객체 영역으로 분류하며, 상기 변화 이미지 객체 영역에 대하여 영상 학습을 수행할 대상으로 추출하는 변화 객체 추출부; 및The image region of the remaining frame is classified into a change image object region composed of change macroblocks and a still image object region composed of still macroblocks, and extracted as an object for performing image learning on the change image object region. Change object extraction unit; And특정 이미지 객체의 중요도가 기준치 이상이면 주요 이미지 객체로 선택하고, 잔여 프레임의 이미지 영역 중 주요 이미지 객체 영역을 영상학습을 수행할 대상으로 추출하되, 주요 객체 추출부;를 포함하고,If the importance of a specific image object is greater than the reference value, and selected as the main image object, and extracts the main image object region of the image area of the remaining frame as a target for the image learning, main object extraction unit;상기 이미지 객체의 중요도는 이미지 객체의 사이즈, 영상 복잡도, 인물 여부 중 적어도 하나를 포함하는 기준에 의해 결정되는 것을 특징으로 하는 동영상 화질 개선 시스템.The importance of the image object is a video quality improvement system, characterized in that determined by a criterion including at least one of the size of the image object, image complexity, whether the person.
- 제 1항에 있어서,The method of claim 1,상기 AI 영상 처리 서버는The AI image processing server영상 학습 결과로 생성된 다운스케일링 파일을 신경망 파일에 기반하여 업스케일링하는 방식의 영상 복원 동작을 수행하되, Performs an image restoration operation of upscaling downscaling files generated as a result of image learning based on neural network files,키 프레임에 대하여 전체 이미지 영역에 대한 영상 복원을 실시하고, 잔여 프레임은 일부 이미지 영역에 대한 영상복원을 수행하고, 나머지 영역은 상기 키프레임의 복원 이미지로부터 붙여 넣는 방식으로 복원 동작을 수행하는 AI 영상 복원부;를 포함하는 것을 특징으로 하는 동영상 화질 개선 시스템.Restoring an image of the entire image region with respect to the key frame, restoring the image with respect to some image regions, and restoring the rest region by pasting from the reconstructed image of the keyframe. Restoring unit; video quality improvement system comprising a.
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR20180047305 | 2018-04-24 | ||
KR10-2018-0047305 | 2018-04-24 | ||
KR10-2019-0046957 | 2019-04-22 | ||
KR1020190046957A KR102179436B1 (en) | 2018-04-24 | 2019-04-22 | Video resolution enhancement system using change macro-block extraction technique |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019209008A1 true WO2019209008A1 (en) | 2019-10-31 |
Family
ID=68294160
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2019/004895 WO2019209008A1 (en) | 2018-04-24 | 2019-04-23 | System for improving video quality by using changed macroblock extraction technique |
Country Status (1)
Country | Link |
---|---|
WO (1) | WO2019209008A1 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111160091A (en) * | 2019-11-22 | 2020-05-15 | 张鸿 | Supply chain intelligent image comparison system and method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110255605A1 (en) * | 1997-05-16 | 2011-10-20 | Shih-Fu Chang | Methods and architecture for indexing and editing compressed video over the world wide web |
WO2015151106A1 (en) * | 2014-04-03 | 2015-10-08 | Chess Vision Ltd | Vision system for monitoring board games and method thereof |
KR101702925B1 (en) * | 2015-05-29 | 2017-02-09 | 주식회사 칩스앤미디어 | An apparatus for scaling a resolution using an image patch of multi video frames and method for using it |
US20170085863A1 (en) * | 2015-09-17 | 2017-03-23 | Legend3D, Inc. | Method of converting 2d video to 3d video using machine learning |
US20170347061A1 (en) * | 2015-02-19 | 2017-11-30 | Magic Pony Technology Limited | Machine Learning for Visual Processing |
-
2019
- 2019-04-23 WO PCT/KR2019/004895 patent/WO2019209008A1/en active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110255605A1 (en) * | 1997-05-16 | 2011-10-20 | Shih-Fu Chang | Methods and architecture for indexing and editing compressed video over the world wide web |
WO2015151106A1 (en) * | 2014-04-03 | 2015-10-08 | Chess Vision Ltd | Vision system for monitoring board games and method thereof |
US20170347061A1 (en) * | 2015-02-19 | 2017-11-30 | Magic Pony Technology Limited | Machine Learning for Visual Processing |
KR101702925B1 (en) * | 2015-05-29 | 2017-02-09 | 주식회사 칩스앤미디어 | An apparatus for scaling a resolution using an image patch of multi video frames and method for using it |
US20170085863A1 (en) * | 2015-09-17 | 2017-03-23 | Legend3D, Inc. | Method of converting 2d video to 3d video using machine learning |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111160091A (en) * | 2019-11-22 | 2020-05-15 | 张鸿 | Supply chain intelligent image comparison system and method |
CN111160091B (en) * | 2019-11-22 | 2023-06-02 | 上海华瑞银行股份有限公司 | Intelligent image comparison system and method for supply chain |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
KR102179436B1 (en) | Video resolution enhancement system using change macro-block extraction technique | |
EP2908547B1 (en) | Information-processing device, information-processing system, information-processing program, and moving image data transmission/reception method adapting the frame rate of moving image data based on frame similarity | |
JP5047740B2 (en) | System and method for creating trick play video stream from compressed normal play video bitstream | |
EP2999232A1 (en) | Media playing method, device and system | |
WO2016010229A1 (en) | Operating method of client and server for streaming service | |
US9153127B2 (en) | Video transmitting apparatus, video receiving apparatus, and video transmission system | |
CN109155840B (en) | Moving image dividing device and monitoring method | |
JP7471725B2 (en) | IMAGE PROCESSING METHOD, APPARATUS, ELECTRONIC DEVICE, AND COMPUTER PROGRAM | |
WO2015060638A1 (en) | Adaptive real-time transcoding method and streaming server therefor | |
CN111405339B (en) | Split screen display method, electronic equipment and storage medium | |
CN115250356A (en) | Multi-camera switchable virtual camera of mobile phone | |
JP2024518227A (en) | Data processing method, device, equipment and computer program | |
EP3399713B1 (en) | Device, system, and method to perform real-time communication | |
WO2019209008A1 (en) | System for improving video quality by using changed macroblock extraction technique | |
RU2697702C1 (en) | Image compression method, image recovery method, image compression device and image recovery device | |
RU2655662C1 (en) | Image processing apparatus and image processing method | |
JP2012137900A (en) | Image output system, image output method and server device | |
WO2019209007A1 (en) | Ai-based image compression and decompression system | |
WO2019209005A1 (en) | Artificial intelligence based resolution improvement system | |
CN110401835B (en) | Image processing method and device | |
WO2019209006A1 (en) | Method for improving resolution of streaming files | |
JP6284172B1 (en) | Screen image transfer method, image restoration method, screen image transfer system, image restoration system, screen image transfer program, image restoration program, image compression method, image compression system, image compression program | |
CN110139144A (en) | One kind is based on TV sharing method in smart home | |
WO2017203789A1 (en) | Difference image generation method, image restoration method, difference detection device, image restoration device, and monitoring method | |
KR102110502B1 (en) | transcoder device for PoE cameras by use of syntax-based object Region-Of-Interest compression |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19793562 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 19793562 Country of ref document: EP Kind code of ref document: A1 |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 19.04.2021) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 19793562 Country of ref document: EP Kind code of ref document: A1 |