KR102364762B1 - Method for training model for super resolution imaging - Google Patents

Abstract

According to one embodiment of the present disclosure, disclosed is a deep learning-based super resolution (SR) imaging method, which is performed by a computing device including at least one processor. The method may include the steps of: receiving a first image and a second image at a lower resolution relative to the first image; and matching characteristics regarding a content density between the first image and the second image. In this case, the content of the first image and the content of the second image may be unpaired.

Description

METHOD FOR TRAINING MODEL FOR SUPER RESOLUTION IMAGING

The present invention relates to an image model learning method, and more particularly, to a deep learning technique for performing super-resolution (SR) imaging using an unpaired low-resolution image and a high-resolution image.

Super resolution (SR) imaging relates to a technique for reconstructing a low resolution image to a high resolution. With the development of deep learning technology, the technical performance of super-resolution imaging is also greatly improved. Most of the deep learning-based super-resolution imaging methods currently being developed are performed by training a model using a paired image set. The paired image set is an image set that is correlated with each other including corresponding contents, and refers to an image set that can be naturally matched or combined.

Since it is very difficult to obtain a paired image set according to a real domain environment, training a model using the paired image set is impractical. For example, in the case of satellite imagery, it is not easy to obtain a matching image set compared to other domains due to numerous environmental and technical factors that affect the acquisition of satellite imagery. Therefore, in the case of satellite imagery, the conventional deep learning-based super-resolution imaging method cannot be applied as it is, and even if it is applied, its performance cannot be guaranteed. Since it is difficult to apply the existing method in an environment where there is a shortage of paired image sets, there is an increasing demand in the art for a technology for performing super-resolution imaging even with an unpaired image set.

Korean Patent Laid-Open No. 10-2018-0120478 (2018.11.06) discloses a method for learning a relationship between domains based on a generative adversarial network.

The present disclosure has been made in response to the above-mentioned background technology, and it is an object of the present disclosure to provide a method of performing super-resolution imaging using an unpaired low-resolution image and a high-resolution image based on deep learning. do it with

A deep learning-based super-resolution (SR) imaging method performed by a computing device according to an embodiment of the present disclosure in order to realize the above-described problem is disclosed. The method comprises: a model learning method for super-resolution imaging performed by a computing device including at least one processor, comprising: receiving a first image and a second image having a lower resolution compared to the first image; and matching a characteristic regarding a contents density between the first image and the second image. In this case, the content of the first image and the content of the second image may be unpaired.

In an alternative embodiment, the matching of the characteristics regarding the content density between the first image and the second image may include extracting a patch of each of the first image and the second image.

In an alternative embodiment, the step of matching the characteristic regarding the content density between the first image and the second image may include deriving a characteristic value of each of the first patch of the first image and the second patch of the second image. step; and selecting the first patch corresponding to the content density-related characteristic of the second patch based on the characteristic value of the first patch and the characteristic value of the second patch.

In an alternative embodiment, the selecting of the first patch corresponding to the content density characteristic of the second patch based on the characteristic value of the first patch and the characteristic value of the second patch may include: calculating a difference between a feature value of , and a feature value of the second patch; and selecting the first patch corresponding to the content density-related characteristic of the second patch based on the difference between the calculated feature values.

In an alternative embodiment, the step of selecting the first patch corresponding to the characteristic related to the content density of the second patch based on the difference in the calculated characteristic values may include comparing the calculated characteristic value difference with a threshold value. to do; and selecting a first patch that satisfies a predetermined condition as a patch corresponding to a content density characteristic of the second patch, based on a result of the comparison.

In an alternative embodiment, the predetermined condition may include a characteristic regarding a content density between the first patch and the second patch when a difference between the characteristic value of the first patch and the characteristic value of the second patch is less than the threshold value. It may include a condition for determining that it corresponds.

In an alternative embodiment, the feature value may include a variance value, which is one of statistical indicators indicating a feature related to a content density of an image.

in an alternative embodiment, training the first model to generate a low-resolution image corresponding to the second image based on the first image in which the characteristic regarding content density is matched with the second image; and based on the output image of the first model, training the second model to generate a high-resolution image corresponding to the first image in which the characteristic related to the content density is matched with the second image. .

A computer program stored in a computer readable storage medium is disclosed according to an embodiment of the present disclosure for realizing the above-described tasks, wherein the computer program performs learning of a model for super-resolution imaging when executed in one or more processors to perform the following operations, wherein the operations include: receiving a second image having a lower resolution than the first image; and matching characteristics related to content density between the first image and the second image.

A computing device for performing super-resolution imaging based on deep learning according to an embodiment of the present disclosure for realizing the above-described problem is disclosed. The apparatus includes: a processor including at least one core; a memory containing program codes executable by the processor; and a network unit for receiving an image, wherein the processor may match a characteristic regarding content density between the first image and the second image having a lower resolution compared to the first image.

According to an embodiment of the present disclosure, there is disclosed a computer-readable recording medium storing a data structure corresponding to a parameter of a neural network that is at least partially updated in a learning process. The operation of the neural network is based at least in part on the parameter, and the method of processing the data includes: receiving a first image and a second image having a lower resolution compared to the first image; and matching characteristics regarding content density between the first image and the second image.

The present disclosure uses an unpaired low resolution (LR: Low Resolution) image and a high-resolution (HR: High Resolution) image based on deep learning to train a model for super-resolution (SR) imaging. method can be provided.

1 is a block diagram of a computing device that performs super-resolution imaging based on deep learning according to an embodiment of the present disclosure.
2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure;
3 is a block diagram illustrating a learning and inference process for generating a super-resolution image of a computing device according to an embodiment of the present disclosure.
4 is a block diagram showing the structure of a deep learning model for performing super-resolution imaging according to an embodiment of the present disclosure.
5 is a flowchart illustrating a process of learning a deep learning model of a computing device according to an embodiment of the present disclosure.
6 is a flowchart illustrating a process of matching characteristics between unpaired images of a computing device according to an embodiment of the present disclosure.
7 is a schematic diagram of a computing environment according to an embodiment of the present disclosure;

Various embodiments are now described with reference to the drawings. In this specification, various descriptions are presented to provide an understanding of the present disclosure. It is evident that the embodiments may be practiced without specific description.

The terms “component,” “module,” “system,” and the like, as used herein, refer to a computer-related entity, hardware, firmware, software, a combination of software and hardware, or execution of software. For example, a component can be, but is not limited to being, a process running on a processor, a processor, an object, a thread of execution, a program, and/or a computer. For example, both an application running on a computing device and the computing device may be a component.

One or more components may reside within a processor and/or thread of execution. A component may be localized within one computer. A component may be distributed between two or more computers. Also, these components may execute from various computer readable media having various data structures stored therein. Components may communicate via a network such as the Internet with another system, for example, via a signal having one or more data packets (eg, data and/or signals from one component interacting with another component in a local system, distributed system, etc.) may communicate via local and/or remote processes depending on the data being transmitted).

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from context, "X employs A or B" is intended to mean one of the natural implicit substitutions. That is, X employs A; X employs B; or when X employs both A and B, "X employs A or B" may apply to either of these cases. It is also to be understood that the term “and/or” as used herein refers to and includes all possible combinations of one or more of the listed related items.

Also, the terms "comprises" and/or "comprising" should be understood to mean that the feature and/or element in question is present. However, it should be understood that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, elements and/or groups thereof. Also, unless otherwise specified or indicating a singular form, the singular in the specification and claims should generally be construed to mean “one or more” unless the context clearly indicates.

And, the term "at least one of A or B" should be interpreted as meaning "when only A is included, when only B is included, when combined with the configuration of A and B".

Those skilled in the art will further appreciate that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, computer software, or combinations of both. It should be recognized that they can be implemented with To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

Descriptions of the presented embodiments are provided to enable those skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art of the present disclosure. The generic principles defined herein may be applied to other embodiments without departing from the scope of the disclosure. Therefore, the present invention is not limited to the embodiments presented herein. This invention is to be construed in its widest scope consistent with the principles and novel features presented herein.

In the present disclosure, a network function, an artificial neural network, and a neural network may be used interchangeably.

On the other hand, the terms “image”, “image” or “image data” used throughout the detailed description and claims of the present disclosure refer to multidimensional data composed of discrete image elements (eg, pixels in a two-dimensional image). In other words, it is a term that refers to a visible object (eg, displayed on a video screen) or a digital representation of that object (eg, a file corresponding to the pixel output of the back).

The term “Nth (N is a natural number)” expressed throughout the detailed description and claims of the present disclosure may be understood as an expression used to distinguish elements of the invention from a functional or structural point of view. The term “Nth” may be understood as a term used to distinguish and express components according to a predetermined criterion. For example, in the present disclosure, two models that are based on the same neural network structure but perform different functions due to differences in input/output data, etc. can be distinguished through an expression called an N-th configuration, like the first model and the second model. In addition, two models that perform the same function but are based on different neural network structures can be distinguished through the expression of the N-th configuration like the first model and the second model. However, this is only an example, and the term “Nth” may be used to distinguish components according to various criteria from various viewpoints.

In addition, the term “NM (N, M is a natural number)” expressed throughout the detailed description and claims of the present disclosure distinguishes elements associated with a specific configuration (ie, N-th configuration) of the invention from each other from a functional or structural point of view. It can be understood as an expression used for For example, the components of the neural network unit included in the first model are distinguished through the expression of the NM configuration, such as the 1-1 module (or neural network), the 1-2 module (or neural network), etc. according to functions or structures. can be However, this is only an example, and the term “N-M” may be used to distinguish components according to various criteria from various viewpoints.

1 is a block diagram of a computing device that performs super-resolution (SR) imaging based on deep learning according to an embodiment of the present disclosure.

The configuration of the computing device 100 shown in FIG. 1 is only a simplified example. In an embodiment of the present disclosure, the computing device 100 may include other components for performing a computing environment of the computing device 100 , and only some of the disclosed components may constitute the computing device 100 .

The computing device 100 may include a processor 110 , a memory 130 , and a network unit 150 .

The processor 110 may include one or more cores, and may include a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU) of a computing device. Unit), etc., may include a processor for data analysis and deep learning. The processor 110 may read a computer program stored in the memory 130 to perform data processing for machine learning according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the processor 110 may perform an operation for learning the neural network. The processor 110 is configured to process input data for learning in deep learning (DL), extract features from input data, calculate an error, and update the weight of the neural network using backpropagation for learning of the neural network. calculations can be performed. At least one of a CPU, a GPGPU, and a TPU of the processor 110 may process learning of a network function. For example, the CPU and the GPGPU may learn a network function together and process data classification using the network function. Also, in an embodiment of the present disclosure, learning of a network function and data classification using the network function may be processed by using the processors of a plurality of computing devices together. In addition, the computer program executed in the computing device according to an embodiment of the present disclosure may be a CPU, GPGPU or TPU executable program.

According to an embodiment of the present disclosure, the processor 110 may perform super-resolution imaging based on an unpaired image set. In this case, the unpaired image set may be understood as including a plurality of images independent of each other including contents that do not correspond to each other. The unmatched image set may be understood to mean an image set in which natural feature matching or combination is difficult due to non-corresponding content. For example, an unpaired image set may include satellite images of different regions. The processor 110 may perform super-resolution imaging of converting a low-resolution image into a high-resolution image by using a deep learning model based on an image set consisting of satellite images captured in different regions.

The processor 110 may perform preprocessing for generating input data of the deep learning model for super-resolution imaging based on the unpaired low-resolution image and the high-resolution image. Content inconsistency of unpaired images inevitably degrades the learning performance for super-resolution imaging of deep learning models. In other words, since unpaired images have different contents, it inevitably takes a lot of time and money to train the deep learning model so that the deep learning model can perform super-resolution imaging normally. Therefore, in order to further improve the learning performance of the deep learning model for super-resolution imaging, the processor 110 may perform preprocessing for matching data characteristics between the low-resolution image and the high-resolution image. That is, the processor 110 may solve the learning difficulty due to content mismatch and increase the learning efficiency through preprocessing that limits the content difference between the low-resolution image and the high-resolution image to a certain level.

For example, the processor 110 may match a characteristic regarding content density between a low-resolution satellite image of a forest and a high-resolution satellite image of a city. The processor 110 may extract a feature value related to each content density of a part of a low-resolution satellite image of a forest and a part of a high-resolution satellite image of a city. The processor 110 may select a part of a high-resolution satellite image of a city corresponding to a part of a low-resolution satellite image of a forest by comparing the extracted feature values regarding the content density. In this way, the processor 110 limits the content difference between the images by matching the characteristics related to the content density between a part of a low-resolution satellite image with different contents and a part of a high-resolution satellite image of a deep learning model for super-resolution imaging. It can improve learning performance.

The processor 110 may train a deep learning model for performing super-resolution imaging based on a plurality of images having data characteristics matched through the above-described pre-processing. The processor 110 may train a deep learning model that makes super-resolution imaging based on a low-resolution image and a high-resolution image in which the content does not correspond to each other, but the content difference is limited to a certain level through preprocessing. In this case, the resolution of the plurality of images may be divided according to an absolute standard resolution that is generally determined according to the number of pixels, or may be divided by a relative difference. For example, the processor 110 learns a deep learning model that has content that does not correspond to each other but performs a super-resolution image based on a part of a low-resolution satellite image and a part of a high-resolution satellite image in which characteristics related to content density are matched. can do it The deep learning model trained by the processor 110 may include a plurality of generative adversarial network (GAN)-based models. A plurality of generative adversarial neural network-based models included in the deep learning model are connected end-to-end, and can be trained to generate super-resolution images based on low-resolution satellite images and high-resolution images. there is.

According to an embodiment of the present disclosure, the memory 130 is a flash memory type, a hard disk type, a multimedia card micro type, and a card type memory (eg, For example, SD or XD memory), RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read Memory) -Only Memory), a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium. The computing device 100 may operate in relation to a web storage that performs a storage function of the memory 130 on the Internet. The description of the above-described memory is only an example, and the present disclosure is not limited thereto.

The network unit 150 according to an embodiment of the present disclosure may use any type of known wired/wireless communication system.

The network unit 150 may receive images from an external system. For example, the network unit 150 may receive ground-captured images from a satellite system, an air system, or the like. The ground-captured images may be data for training or inference of a neural network model. Also, the terrestrial photographed images may be images that are mutually paired or images that are not paired. The ground-captured images may include both an electro-optic image and a Synthetic Aperture Radar (SAR) image captured through an artificial satellite, an aircraft, or the like. The ground-captured images are not limited to the above-described examples, and may be variously configured within a range that those skilled in the art can understand.

Also, the network unit 150 may transmit/receive information and data processed by the processor 110 through communication with other terminals. For example, the network unit 150 may provide the super-resolution image generated by the processor 110 to the client (e.g. other servers, user terminals, etc.).

Meanwhile, the computing device 100 according to an embodiment of the present disclosure may include a server as a computing system that transmits and receives information through communication with a client. In this case, the client may be any type of terminal capable of accessing the server. For example, the computing device 100, which is a server, performs super-resolution imaging by receiving unpaired terrestrial images from the satellite system, and image analysis (eg object detection, etc.) on the super-resolution image generated as a result of the operation. It can be provided as a server or user terminal for

In an additional embodiment, the computing device 100 may include any type of terminal that receives a data resource generated from an arbitrary server and performs additional information processing.

2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure;

A model according to an embodiment of the present disclosure may include a neural network for performing super-resolution imaging. Throughout this specification, the terms neural network, network function, and neural network may be used interchangeably. A neural network may be composed of a set of interconnected computational units that can be generally referred to as nodes. These nodes may also be referred to as neurons. A neural network is configured to include at least one or more nodes. Nodes or neurons constituting the neural networks may be interconnected by one or more links.

In the neural network, one or more nodes connected through a link may relatively form a relationship between an input node and an output node. The concepts of an input node and an output node are relative, and any node in an output node relationship with respect to one node may be in an input node relationship in a relationship with another node, and vice versa. As described above, an input node-to-output node relationship may be created around a link. One or more output nodes may be connected to one input node through a link, and vice versa.

In the relationship between the input node and the output node connected through one link, the value of the data of the output node may be determined based on data input to the input node. Here, a link interconnecting the input node and the output node may have a weight. The weight may be variable, and may be changed by a user or an algorithm in order for the neural network to perform a desired function. For example, when one or more input nodes are interconnected to one output node by respective links, the output node sets values input to input nodes connected to the output node and links corresponding to the respective input nodes. An output node value may be determined based on the weight.

As described above, in a neural network, one or more nodes are interconnected through one or more links to form an input node and an output node relationship in the neural network. The characteristics of the neural network may be determined according to the number of nodes and links in the neural network, the correlation between the nodes and the links, and the value of a weight assigned to each of the links. For example, when the same number of nodes and links exist and there are two neural networks having different weight values of the links, the two neural networks may be recognized as different from each other.

A neural network may consist of a set of one or more nodes. A subset of nodes constituting the neural network may constitute a layer. Some of the nodes constituting the neural network may constitute one layer based on distances from the initial input node. For example, a set of nodes having a distance of N from an initial input node may constitute N layers. The distance from the initial input node may be defined by the minimum number of links that must be passed to reach the corresponding node from the initial input node. The definition of such a layer is arbitrary for explanation, and the order of the layer in the neural network may be defined in a different way than the above-described method. For example, a layer of nodes may be defined by a distance from the final output node.

The initial input node may mean one or more nodes to which data is directly input without going through a link in a relationship with other nodes among nodes in the neural network. Alternatively, in a relationship between nodes based on a link in a neural network, it may mean nodes that do not have other input nodes connected by a link. Similarly, the final output node may refer to one or more nodes that do not have an output node in relation to other nodes among nodes in the neural network. In addition, the hidden node may mean nodes constituting the neural network other than the first input node and the last output node.

The neural network according to an embodiment of the present disclosure may have the same number of nodes in the input layer as the number of nodes in the output layer, and may be a neural network in which the number of nodes decreases and then increases again as the input layer progresses from the input layer to the hidden layer. there is. In addition, in the neural network according to another embodiment of the present disclosure, the number of nodes in the input layer is greater than the number of nodes in the output layer. It may be small, and it may be a neural network in which the number of nodes decreases as it progresses from the input layer to the hidden layer. In addition, in the neural network according to another embodiment of the present disclosure, the number of nodes in the input layer is higher than the number of nodes in the output layer. There may be many, and it may be a neural network in which the number of nodes increases as the input layer progresses to the hidden layer. The neural network according to another embodiment of the present disclosure may be a neural network in a combined form of the aforementioned neural networks.

A deep neural network (DNN) may refer to a neural network including a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks can be used to identify the latent structures of data. In other words, it can identify the potential structure of images, text, video, voice, and music (e.g., what objects are in the image, what the text and emotions are, what the content and emotions of speech are, etc.) . Deep neural networks are Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Auto Encoder, Restricted Boltzmann Machine (RBM), Deep Trust Network ( DBN: Deep Belief Network), a Q network, a U network, a Siamese network, a Generative Adversarial Network (GAN), etc. may be included. The description of the deep neural network described above is only an example, and the present disclosure is not limited thereto.

In an embodiment of the present disclosure, the network function may include an autoencoder. The auto-encoder may be a kind of artificial neural network for outputting output data similar to input data. The auto encoder may include at least one hidden layer, and an odd number of hidden layers may be disposed between the input/output layers. The number of nodes in each layer may be reduced from the number of nodes of the input layer to an intermediate layer called the bottleneck layer (encoding), and then expanded symmetrically with reduction from the bottleneck layer to the output layer (symmetrically with the input layer). The auto-encoder can perform non-linear dimensionality reduction. The number of input layers and output layers may correspond to a dimension after preprocessing the input data. In the auto-encoder structure, the number of nodes of the hidden layer included in the encoder may have a structure that decreases as the distance from the input layer increases. If the number of nodes in the bottleneck layer (the layer with the fewest nodes located between the encoder and decoder) is too small, a sufficient amount of information may not be conveyed, so a certain number or more (e.g., more than half of the input layer, etc.) ) may be maintained.

The neural network may be trained by at least one of teacher learning, unsupervised learning, semi-supervised learning, and reinforcement learning. Learning of the neural network may be a process of applying knowledge for the neural network to perform a specific operation to the neural network.

A neural network can be trained in a way that minimizes output errors. In the training of a neural network, iteratively inputs the training data to the neural network, calculates the output of the neural network and the target error for the training data, and calculates the error of the neural network from the output layer of the neural network to the input layer in the direction to reduce the error. It is a process of updating the weight of each node in the neural network by backpropagation in the direction. In the case of teacher learning, learning data in which the correct answer is labeled in each learning data is used (ie, labeled learning data), and in the case of comparative learning, the correct answer may not be labeled in each learning data. That is, for example, learning data in the case of teacher learning related to data classification may be data in which categories are labeled in each of the learning data. The labeled training data is input to the neural network, and an error can be calculated by comparing the output (category) of the neural network with the label of the training data. As another example, in the case of comparative learning about data classification, an error may be calculated by comparing the input training data with the neural network output. The calculated error is back propagated in the reverse direction (ie, from the output layer to the input layer) in the neural network, and the connection weight of each node of each layer of the neural network may be updated according to the back propagation. The change amount of the connection weight of each node to be updated may be determined according to a learning rate. The computation of the neural network on the input data and the backpropagation of errors may constitute a learning cycle (epoch). The learning rate may be applied differently according to the number of repetitions of the learning cycle of the neural network. For example, in the early stage of learning of a neural network, a high learning rate can be used to enable the neural network to quickly obtain a certain level of performance, thereby increasing efficiency, and using a low learning rate at a later stage of learning can increase accuracy.

In the training of neural networks, in general, the training data may be a subset of real data (that is, data to be processed using the trained neural network), and thus, the error on the training data is reduced, but the error on the real data is reduced. There may be increasing learning cycles. Overfitting is a phenomenon in which errors on actual data increase due to over-learning on training data. For example, a phenomenon in which a neural network that has learned a cat by seeing a yellow cat does not recognize that it is a cat when it sees a cat other than yellow may be a type of overfitting. Overfitting can act as a cause of increasing errors in machine learning algorithms. In order to prevent such overfitting, various optimization methods can be used. In order to prevent overfitting, methods such as increasing the training data, regularization, dropout that deactivate some nodes of the network in the process of learning, and the use of a batch normalization layer can be applied.

3 is a block diagram illustrating a learning and inference process for generating a super-resolution image of a computing device according to an embodiment of the present disclosure.

Referring to FIG. 3 , the processor 110 of the computing device 100 according to an embodiment of the present disclosure evaluates data characteristics between input data in order to efficiently train a deep learning model generating a super-resolution image 13 . Matching preprocessing can be performed. The processor 110 may perform an operation for matching data characteristics between the first image 11 and the second image 12 having a lower resolution compared to the first image 11 . The content of the first image 11 and the content of the second image 12 may not correspond to each other. For example, the first image 11 may include a satellite image of a specific area. The second image 12 is a satellite image obtained by photographing an area different from the first image 11 , and may include a relatively low-resolution image compared to the first image 11 . As such, the first image 11 and the second image 12 may be images in which elements constituting content, such as an object, a background, and a style, are different. The difference in resolution between the first image 11 and the second image 12 may be based on a relative comparison between images, or based on a resolution standard (eg HD: High Definition, etc.) determined according to the number of pixels. it may be

The processor 110 may use the matching module 220 to match characteristics related to content density between the high-resolution first image 11 and the low-resolution second image 12 . The processor 110 generates at least a portion of the first image 11 for training of the deep learning model based on the characteristic regarding the content density between at least a portion of the first image 11 and at least a portion of the second image 12 . can be selected Here, the content density may be determined using statistical characteristics of the image. The content density may be understood as an index indicating the proportion of meaningful content using image statistics obtained from a partial region of an image.

For example, in an image, image statistics for an area containing a cloudless sky and an area containing a human face may be different. Also, in an image, the content density of an area containing a human face may be higher than that of an area containing the sky. As can be seen from the above example, the processor 110 may estimate the content density using image statistics of a part of an image (or image patch), and may obtain two patches having similar content densities using this.

Content density may be understood as an index indicating statistical characteristics of an image based on a specific region of the image. The content density may be understood as an index indicating the proportion of meaningful content in a specific area of an image.

For example, in an image of a city center with a satellite orbiting a fixed orbit, there may be an area that contains only buildings, an area where buildings are partially covered by clouds, and an area where no objects are identified because all objects are obscured by clouds. there is. Assuming that patches of the same size are extracted and compared for each area, the content density may vary for each patch depending on the degree to which the building content is covered by clouds. It can be seen that a patch in an area containing only buildings has a higher content density than a patch in an area in which buildings are partially covered by clouds. Also, it can be seen that the patch in the area where the building is partially covered by the cloud has a higher content density than the patch in the area in which all objects are covered by the cloud. As can be seen from the above example, the content density may be understood as indicating how high-purity content is included in an image based on an area of the same size.

The processor 110 may extract each patch of the first image 11 and the second image 12 in order to match the characteristics regarding the content density between the first image 11 and the second image 12 . . The processor 110 may extract patches of corresponding sizes from each of the first image 11 and the second image 12 . Since the contents of the first image 11 and the second image 12 do not correspond to each other, it is difficult to match the characteristics of the content density with the image itself. Therefore, according to the purpose of limiting the difference between unpaired images as much as possible so that efficient learning can be performed, the processor 110 extracts a patch of each of the images 11 and 22, and the content density in the patch unit It is possible to perform a task of matching the characteristics of . In this case, the patch may be a set of pixels constituting the image, or may be at least a part of the image extracted with a predetermined size according to a predetermined criterion.

The processor 110 is configured to match a characteristic regarding content density between the first image 11 and the second image 12 , a patch (hereinafter, a first patch) of the first image 11 and a second image 12 . ) of each patch (hereinafter, the second patch) may be derived. The processor 110 may derive a feature value representing a characteristic regarding the content density of each of the first patch and the second patch. For example, unpaired first image 11 and second image 12 are independent of each other, and thus may contain completely unrelated content as well as varying degrees of image statistics. That is, since the two images 11 and 12 are not paired, a difference in the degree of statistical values occurs. By limiting the difference, the characteristics regarding the content density can be matched. Accordingly, the processor 110 may extract statistical values such as variance values and average values of the contents in the patch as feature values of each patch and use them to match the characteristics related to the content density. Since the detailed description of the above-described characteristic values is only one example, the types of characteristic values may be variously configured within a range that can be understood by those skilled in the art based on the above-described examples.

The processor 110 may select the first patch corresponding to the characteristic regarding the content density of the second patch based on the characteristic value of the first patch and the characteristic value of the second patch. The processor 110 may compare the feature value of the first patch with the feature value of the second patch to determine the first patch that matches the content density feature with the second patch. That is, matching the characteristics regarding the content density between the first image 11 and the second image 12 means that at least a portion of the first image 11 to which the characteristics regarding the content density and the second image 12 correspond to each other. It can be understood as a selection operation. For example, the processor 110 compares the statistical value of the first patch with the statistical value of the second patch, and selects the first patch with a small difference from the statistical value of the second patch as the training data of the deep learning model. there is. The processor 110 uses, as training data, the first patch matching the characteristics related to the second patch and the content density through a comparison operation between the statistical value of the first patch and a predetermined threshold value based on the statistical value of the second patch. can decide

The processor 110 may train the first model 300 that performs resolution conversion of the image based on the first image and the second image 12 in which the second image 12 and the characteristics related to the content density are matched. there is. The processor 110 inputs the patch of the first image and the patch of the second image 12, in which the patch of the second image 12 and the characteristic related to the content density are matched, to the first model 300 to input the second image ( The first model 300 may be trained to generate a low-resolution image corresponding to 12). For example, the processor 110 may train the first model 300 to decrease the resolution of the first patch by using the second patch as a ground truth (GT). It may be understood by those skilled in the art that the learning of the first model 300 is performed to well estimate a function for resolution reduction expressed as in Equation 1 below.

Here, k denotes a kernel, n denotes noise, and s denotes a scale factor. Since the values of k and n are unknown, (x*k)↓ _s corresponds to the deep-stacked neural network so that the first model 300 performs an operation corresponding to the above-described function, so that the processor 110 is the first model The 300 may be taught to adjust the resolution of the first patch. In other words, the first model 300 may be trained to convert the first patch to a low resolution corresponding to the second patch based on the first patch and the second patch. At this time, since the content features of the first image 11 and the second image 12 do not correspond to each other, a model having a general neural network structure that matches the features by performing GT (ground truth) on the second image 12 is used. It is difficult to use as the first model 300 . Therefore, in order to solve the above-mentioned problem, in the first model 300 of the present disclosure, the first patch is the input of the generator and the second patch is the GT (ground truth) of the discriminator. An adversarial neural network-based model may be used.

The processor 110 may train the second model 400 that performs resolution conversion of an image based on the output image of the first model 300 and the first image 11 . The processor 110 inputs the output image and the first image 11 of the first model 300 to the second model 400 to generate a high-resolution image corresponding to the first image 11 ( 400) can be learned. For example, the processor 110 may train the second model 400 to improve the resolution of the output image of the first model 300 based on the first image 11 . In other words, the second model 400 can be learned to convert the low-resolution image corresponding to the second image 12 to high-resolution based on the first image 11 and the output image of the first model 300 . there is. At this time, as the second model 400 of the present disclosure, there is a generative adversarial neural network-based model in which the output image of the first model 300 is the input of the generator and the first image 11 is the ground truth (GT) of the discriminator. can be used

The processor 110 may perform super-resolution imaging on unpaired images using the first model 300 and the second model 400 pre-trained as described above. The processor 110 may generate a super-resolution image paired with the low-resolution image based on the high-resolution image that is not paired with the low-resolution image through the pre-trained first model 300 and the second model 400 . For example, in FIG. 3 , assuming that the first image 11 is a high-resolution input image and the second image 12 is a low-resolution image that is not paired with the first image, pairing with the second image 15 is In order to generate the super-resolution image, the processor 110 may use the first model 300 and the second model 400 pre-trained as described above. The first model 300 may receive the first image 11 and generate a low-resolution image corresponding to the second image 12 . The second model 400 may receive the low-resolution image generated by the first model 300 and generate a super-resolution image 13 having a resolution corresponding to the first image 11 . Through the operation of the processor 110 using the first model 300 and the second model 400, super-resolution imaging that improves the resolution of low-resolution images even with unpaired image sets can be effectively performed. can

4 is a block diagram showing the structure of a deep learning model for performing super-resolution imaging according to an embodiment of the present disclosure.

Referring to FIG. 4 , the matching module 200 included in the processor 110 of the computing device 100 according to an embodiment of the present disclosure is used for learning the first model 300 and the second model 400 . In order to generate input data, at least a portion 23 of the first image 21 corresponding to the characteristic relating to the content density of the second image 22 may be selected.

For example, the matching module 200 may generate a first patch corresponding to at least a portion 23 of the first image 21 and a second patch corresponding to at least a portion of the second image 22 . In this case, each patch may have a size corresponding to each other. The matching module 200 calculates a difference between a feature value of the first patch corresponding to at least a part 23 of the first image 21 and a feature value of a second patch corresponding to at least a part of the second image 22 . can do. For example, the matching module 200 may extract each patch from a high-resolution image of a city center and a low-resolution image of a forest. The matching module 200 may extract, as a feature value, a variance value, which is a statistical value indicating the distribution of content, in a patch of a high-resolution image of a city center. The matching module 200 may extract, as a feature value, a variance value, which is a statistical value indicating the distribution of content in a patch of a low-resolution image of a forest. The matching module 200 may calculate a difference between the variance values of each patch in order to match the characteristics related to the content density between the two images.

The matching module 200 may select the first patch corresponding to the characteristic regarding the content density of the second patch based on a difference between the characteristic value of the first patch and the characteristic value of the second patch. The matching module 200 may compare a difference between the feature value of the first patch and the feature value of the second patch and a threshold value. The matching module 200 may select a first patch that satisfies a predetermined condition as a patch corresponding to the content density characteristic of the second patch based on the result of the comparison. In this case, the predetermined condition may be a condition in which, when the difference between the characteristic value of the first patch and the characteristic value of the second patch is less than a threshold value, it is determined that the characteristic regarding the content density between the first patch and the second patch corresponds. .

For example, the matching module 200 may compare a difference between a variance value of a patch of a high-resolution image photographing a city center and a variance value of a patch of a low-resolution image photographing a forest with a predetermined threshold value. In this case, the threshold value may be determined empirically based on the user's sufficient prior knowledge or may be defined by an arbitrary algorithm. The matching module 200 may determine whether the difference between the variance values is greater than or less than a threshold value. When the difference between the variance values is greater than the threshold value, the matching module 200 may determine that the content density characteristic of the first patch does not correspond to the content density characteristic of the second patch. In other words, when it is determined that the difference in variance value is equal to or greater than the threshold value, the matching module 200 determines that the first patch used for calculation has a large difference in content density from the second patch, and may be excluded from the training data. . Conversely, when the difference between the variance values is smaller than the threshold value, the matching module 200 may determine that the content density characteristic of the first patch corresponds to the content density characteristic of the second patch. In other words, when it is determined that the difference between the variance values is less than the threshold value, the matching module 200 determines that the first patch used for the calculation has a small difference in content density from the second patch and selects it as training data. .

According to an embodiment of the present disclosure, a predetermined condition for selecting the above-described first patch may be expressed as the following [Equation 2].

는 제 1 패치의 분산값,

는 제 2 패치의 분산값,

는 제 1 패치와 제 2 패치의 분산 차이를 제어하는 임계값을 나타낸다. [수학식 2]를 참조하면, 매칭 모듈(200)은 분산값의 차이의 절대값이 임계값 미만인 조건을 만족하는 제 1 패치를 제 2 패치의 콘텐츠 밀도에 관한 특성에 대응되는 패치로 선별한다는 것을 알 수 있다. 한편, 특징값의 예시로 설명한 분산값은 패치에서 표현하는 콘텐츠와 관련도가 높다. 따라서, 분산값이 큰 패치에는 풍부한 콘텐츠가 포함되어 있는 것으로 이해될 수 있다. 다만, 콘텐츠 밀도에 관한 특성을 나타내는 특징값으로 설명한 분산값은 하나의 예시일 뿐이므로, 당업자가 이해할 수 있는 범위에서 다양한 특징값들이 적용될 수 있다. At this time,

is the variance of the first patch,

is the variance of the second patch,

denotes a threshold value controlling the variance difference between the first patch and the second patch. Referring to [Equation 2], the matching module 200 selects the first patch that satisfies the condition that the absolute value of the difference in variance is less than the threshold value as a patch corresponding to the content density characteristic of the second patch. it can be seen that On the other hand, the variance value described as an example of the feature value is highly related to the content expressed in the patch. Accordingly, it can be understood that a patch having a large variance value includes rich content. However, since the dispersion value described as a characteristic value indicating a characteristic related to the content density is only an example, various characteristic values may be applied within a range that can be understood by those skilled in the art.

Referring to FIG. 4 , a first model 300 that converts a high-resolution image into a low-resolution image according to an embodiment of the present disclosure receives a first patch 23 that is at least a part of the first image 21 and receives a second image A 1-1 neural network 310 that generates a low-resolution image corresponding to (22), a low-resolution second neural network 310 generated through the 1-1 neural network 310 using the second image 22 as GT (ground truth) 3 may include a 1-2 neural network 320 for discriminating the image 24 .

The 1-1 neural network 310 may receive the first patch 23 and generate a third image 24 that is a low-resolution image close to the second image 22 . In this case, the first patch 23 may be an image in which the properties related to the content density are matched with the second image 22 input to the first model 300 through the matching module 200 . The 1-1 neural network 310 may convert the resolution of the first patch 23 from a high resolution to a low resolution based on the second image 22 . For example, the 1-1 neural network 310 may be understood as a generator of a generative adversarial neural network that adjusts the resolution of the first patch 23 . Therefore, the 1-1 neural network 310 is close to the resolution of the second image 22 so that the 1-2 neural network 320, which will be described later, makes it difficult to distinguish the second image 22 and the third image 24. It aims to create a third image 24 . That is, the 1-1 neural network 310 learns to generate the third image 24 corresponding to the second image 22 from the first patch 23 through competition with the 1-2 neural network 320 . can be

The 2-1 neural network 410 may receive the third image 24 and generate a fourth image 25 that is a high-resolution image close to the first image 21 . The 2-1 neural network 410 may convert the resolution of the third image 24 from a low resolution to a high resolution based on the first image 21 . For example, the 2-1 th neural network 410 may be understood as a generator of a generative adversarial neural network that adjusts the resolution of the third image 24 . Therefore, the 2-1 neural network 410 is close to the resolution of the first image 21 so that the 2-2 neural network 420 to be described later makes it difficult to distinguish the first image 21 from the fourth image 25 . It aims to create a fourth image 25 . That is, the 2-1 neural network 410 learns to generate the fourth image 25 corresponding to the first image 21 from the third image 24 through competition with the 2-2 neural network 420 . can be

5 is a flowchart illustrating a process of learning a deep learning model of a computing device according to an embodiment of the present disclosure.

Referring to FIG. 5 , in step S110 , the computing device 100 according to an embodiment of the present disclosure may receive an image set including a plurality of images whose contents do not correspond to each other from an external system. In this case, the plurality of images included in the image set may be images having different resolutions. For example, the computing device 100 may receive an image set including a first image and a second image having a lower resolution than the first image from the database server. The image set received by the computing device 100 may be used to train a deep learning model for performing super-resolution imaging in a step to be described later.

In operation S120 , the computing device 100 may match characteristics according to the content density between the first image received in operation S110 and the second image having a lower resolution than the first image. For example, the computing device 100 may extract a first patch and a second patch that are at least a part of an image from the first image and the second image, respectively. In addition, the computing device 100 may extract data feature values of each of the first patch and the second patch in order to match the characteristics related to the content density of the first patch and the second patch. In this case, the feature value may include an average value or a variance value indicating the content density of the image. The computing device 100 may match the first patch and the second patch that satisfy a predetermined condition based on the extracted data feature value. In this case, the predetermined condition based on the feature value is a condition for comparing the difference between the feature values and the threshold value. It may include a condition for selecting the patch as a patch matching the second patch.

In step S130, the computing device 100 may train the first model included in the deep learning model to adjust the resolution of a high-resolution image among a plurality of images whose characteristics related to content density are matched with each other through step S120. . The computing device 100 may train the first model to reduce the resolution of the high-resolution image based on the high-resolution image and the low-resolution image that are not paired with each other but have matching characteristics regarding content density. For example, the computing device 100 receives the patch of the first image in which the first model matches the patch of the second image and the characteristic according to the content density through step S120, and a low-resolution image corresponding to the second image. The first model can be trained to generate In this case, the content of the first image does not correspond to the content of the second image.

In step S140 , the computing device 100 may train the second model included in the deep learning model to adjust the resolution of the low-resolution image generated through the first model. The computing device 100 may train the second model to improve the low-resolution image generated through the first model to high resolution. In this case, the second model may be connected end-to-end with the first model. For example, the computing device 100 may train the second model so that the second model receives an output image of the first model and generates a high-resolution image corresponding to the first image. In other words, the computing device 100 may train the second model to generate a high-resolution image that can be paired with the second image from the output image of the first model to which the second model has received the first image. there is.

6 is a flowchart illustrating a process of matching characteristics between unpaired images of a computing device according to an embodiment of the present disclosure.

Referring to FIG. 6 , in step S210 , the computing device 100 according to an embodiment of the present disclosure receives a first image and a second image that is a relatively low-resolution image compared to the first image in order to perform super-resolution imaging. can do. In this case, the contents of the first image and the second image may not correspond to each other. For example, the first image may be a high-resolution satellite image of a city center. The second image may be a low-resolution satellite image of agricultural land. Since the first image contains buildings and roads in the city center as content, while the second image contains crops, land, etc. of agricultural land as content, the first image and the second image can be viewed as including different content from each other. there is. Accordingly, the first image and the second image may be viewed as unpaired images because the contents do not correspond to each other.

In step S220, for efficient learning of the deep learning model for super-resolution imaging, the computing device 100 may derive a feature value corresponding to each content density from the first image and the second image described in step S210. there is. For example, the computing device 100 may derive a variance value and/or an average value as a feature value corresponding to the content density of the first image and the second image. In this case, the variance value and/or the average value may correspond to the content density of the image. That is, the feature value may be understood as an index indicating how the content density is distributed based on a specific region of an image that is a target for extracting the feature value. The above examples, such as variance values and average values, are examples for helping understanding of the description, and do not limit the scope of the feature values.

In step S230 , the computing device 100 performs at least a part corresponding to the content density characteristic of at least a part of the second image that satisfies a predetermined condition based on the feature value corresponding to the content density derived in step S220 . 1 You can select an image to match the image. For example, the computing device 100 may match a characteristic regarding content density between images by comparing a difference in variance between input images with a threshold value. The computing device 100 may determine that the characteristics related to the content density match the first image and the second image in which the difference between the variance values is less than a threshold value. In this case, the threshold value may be predetermined based on prior knowledge based on the user's experience, or may be defined by an arbitrary algorithm or the like. The computing device 100 may use at least a portion of the first image matched with the second image as an input of a deep learning model for super-resolution imaging together with the second image.

7 is a simplified and general schematic diagram of a computing environment in which embodiments of the present disclosure may be implemented.

Although the present disclosure has been described above as being generally capable of being implemented by a computing device, those skilled in the art will appreciate that this disclosure can be executed on one or more computers in combination with computer-executable instructions and/or other program modules and/or in combination of hardware and software. It will be well known that it can be implemented as

Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will appreciate that the methods of the present disclosure can be applied to single-processor or multiprocessor computer systems, minicomputers, mainframe computers as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and the like. It will be appreciated that each of these may be implemented in other computer system configurations, including those capable of being operated in connection with one or more associated devices.

The described embodiments of the present disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Computers typically include a variety of computer-readable media. Any medium accessible by a computer can be a computer-readable medium, and such computer-readable media includes volatile and nonvolatile media, transitory and non-transitory media, removable and non-transitory media. including removable media. By way of example, and not limitation, computer-readable media may include computer-readable storage media and computer-readable transmission media. Computer readable storage media includes volatile and nonvolatile media, temporary and non-transitory media, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. includes media. A computer-readable storage medium may be RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Video Disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device, or other magnetic storage device. device, or any other medium that can be accessed by a computer and used to store the desired information.

Computer readable transmission media typically embodies computer readable instructions, data structures, program modules or other data, etc. in a modulated data signal such as a carrier wave or other transport mechanism, and Includes all information delivery media. The term modulated data signal means a signal in which one or more of the characteristics of the signal is set or changed so as to encode information in the signal. By way of example, and not limitation, computer-readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer-readable transmission media.

An example environment 1100 implementing various aspects of the present disclosure is shown including a computer 1102 , the computer 1102 including a processing unit 1104 , a system memory 1106 , and a system bus 1108 . do. A system bus 1108 couples system components, including but not limited to system memory 1106 , to the processing device 1104 . The processing device 1104 may be any of a variety of commercially available processors. Dual processor and other multiprocessor architectures may also be used as processing unit 1104 .

The system bus 1108 may be any of several types of bus structures that may further interconnect a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 1106 includes read only memory (ROM) 1110 and random access memory (RAM) 1112 . A basic input/output system (BIOS) is stored in non-volatile memory 1110, such as ROM, EPROM, EEPROM, etc., the BIOS is the basic input/output system (BIOS) that helps transfer information between components within computer 1102, such as during startup. contains routines. RAM 1112 may also include high-speed RAM, such as static RAM, for caching data.

These drives and their associated computer-readable media provide non-volatile storage of data, data structures, computer-executable instructions, and the like. In the case of computer 1102, drives and media correspond to storing any data in a suitable digital format. Although the description of computer readable media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art will use zip drives, magnetic cassettes, flash memory cards, cartridges, etc. It will be appreciated that other tangible computer-readable media, such as, etc., may also be used in the example operating environment and that any such media may contain computer-executable instructions for performing the methods of the present disclosure.

A number of program modules may be stored in the drive and RAM 1112 , including an operating system 1130 , one or more application programs 1132 , other program modules 1134 , and program data 1136 . All or portions of the operating system, applications, modules, and/or data may also be cached in RAM 1112 . It will be appreciated that the present disclosure may be implemented in various commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 1102 via one or more wired/wireless input devices, for example, a keyboard 1138 and a pointing device such as a mouse 1140 . Other input devices (not shown) may include a microphone, IR remote control, joystick, gamepad, stylus pen, touch screen, and the like. Although these and other input devices are connected to the processing unit 1104 through an input device interface 1142 that is often connected to the system bus 1108, parallel ports, IEEE 1394 serial ports, game ports, USB ports, IR interfaces, It may be connected by other interfaces, etc.

A monitor 1144 or other type of display device is also coupled to the system bus 1108 via an interface, such as a video adapter 1146 . In addition to the monitor 1144, the computer typically includes other peripheral output devices (not shown), such as speakers, printers, and the like.

Computer 1102 may be operated in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 1148 via wired and/or wireless communications. Remote computer(s) 1148 may be workstations, computing device computers, routers, personal computers, portable computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, and are typically connected to computer 1102 . Although it includes many or all of the components described for it, only memory storage device 1150 is shown for simplicity. The logical connections shown include wired/wireless connections to a local area network (LAN) 1152 and/or a larger network, eg, a wide area network (WAN) 1154 . Such LAN and WAN networking environments are common in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to a worldwide computer network, for example, the Internet.

When used in a LAN networking environment, the computer 1102 is coupled to the local network 1152 through a wired and/or wireless communication network interface or adapter 1156 . Adapter 1156 may facilitate wired or wireless communication to LAN 1152 , which LAN 1152 also includes a wireless access point installed therein for communicating with wireless adapter 1156 . When used in a WAN networking environment, the computer 1102 may include a modem 1158, be connected to a communication computing device on the WAN 1154, or establish communications over the WAN 1154, such as over the Internet. have other means. A modem 1158 , which may be internal or external and a wired or wireless device, is coupled to the system bus 1108 via a serial port interface 1142 . In a networked environment, program modules described for computer 1102 , or portions thereof, may be stored in remote memory/storage device 1150 . It will be appreciated that the network connections shown are exemplary and other means of establishing a communication link between the computers may be used.

Computer 1102 may be associated with any wireless device or object that is deployed and operates in wireless communication, eg, printers, scanners, desktop and/or portable computers, Portable Data Assistants (PDAs), communication satellites, wireless detectable tags. It operates to communicate with any device or place, and phone. This includes at least Wi-Fi and Bluetooth wireless technologies. Accordingly, the communication may be a predefined structure as in a conventional network or may simply be an ad hoc communication between at least two devices.

Wi-Fi (Wireless Fidelity) makes it possible to connect to the Internet, etc. without a wired connection. Wi-Fi is a wireless technology such as cell phones that allows these devices, eg, computers, to transmit and receive data indoors and outdoors, ie anywhere within range of a base station. Wi-Fi networks use a radio technology called IEEE 802.11 (A, B, G, etc) to provide secure, reliable, and high-speed wireless connections. Wi-Fi can be used to connect computers to each other, to the Internet, and to wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks can operate in unlicensed 2.4 GHz and 5 GHz radio bands, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or in products that include both bands (dual band). there is.

On the other hand, a computer-readable medium storing a data structure according to an embodiment of the present disclosure is disclosed.

The data structure may refer to the organization, management, and storage of data that enables efficient access and modification of data. A data structure may refer to an organization of data to solve a specific problem (eg, data retrieval, data storage, and data modification in the shortest time). A data structure may be defined as a physical or logical relationship between data elements designed to support a particular data processing function. The logical relationship between data elements may include a connection relationship between user-defined data elements. Physical relationships between data elements may include actual relationships between data elements physically stored on a computer-readable storage medium (eg, persistent storage). A data structure may specifically include a set of data, relationships between data, and functions or instructions applicable to data. Through an effectively designed data structure, the computing device can perform an operation while using the resource of the computing device to a minimum. Specifically, the computing device may increase the efficiency of operations, reads, insertions, deletions, comparisons, exchanges, and retrievals through effectively designed data structures.

A data structure may be classified into a linear data structure and a non-linear data structure according to the type of the data structure. The linear data structure may be a structure in which only one piece of data is connected after one piece of data. A linear data structure may include a list, a stack, a queue, and a deque. A list may mean a set of data in which an order exists internally. The list may include a linked list. The linked list may be a data structure in which data is linked in such a way that each data is linked in a line with a pointer. In a linked list, a pointer may contain information about a link with the next or previous data. A linked list may be expressed as a single linked list, a doubly linked list, and a circularly linked list according to a shape. A stack can be a data enumeration structure with limited access to data. A stack can be a linear data structure in which data can be processed (eg, inserted or deleted) only at one end of the data structure. The data stored in the stack may be a data structure (LIFO: Last In First Out) that enters later and comes out earlier. A queue is a data listing structure that can access data with limited access, and unlike a stack, the data stored later may be a data structure that comes out later (FIFO: First In First Out). A deck can be a data structure that can process data at either end of the data structure.

The non-linear data structure may be a structure in which a plurality of data is connected after one data. The non-linear data structure may include a graph data structure. A graph data structure may be defined as a vertex and an edge, and an edge may include a line connecting two different vertices. A graph data structure may include a tree data structure. The tree data structure may be a data structure in which one path connects two different vertices among a plurality of vertices included in the tree. That is, it may be a data structure that does not form a loop in the graph data structure.

Throughout this specification, computational model, neural network, network function, and neural network may be used interchangeably. Hereinafter, the neural network is unified and described. The data structure may include a neural network. And the data structure including the neural network may be stored in a computer-readable medium. Data structures, including neural networks, also include preprocessed data for processing by the neural network, data input to the neural network, weights of the neural network, hyperparameters of the neural network, data obtained from the neural network, activation functions associated with each node or layer of the neural network, and neural networks. It may include a loss function for learning of . A data structure comprising a neural network may include any of the components disclosed above. That is, the data structure including the neural network includes preprocessed data for processing by the neural network, data input to the neural network, weights of the neural network, hyperparameters of the neural network, data obtained from the neural network, activation functions associated with each node or layer of the neural network, It may be configured to include all or any combination thereof, such as a loss function for learning of a neural network. In addition to the above-described configurations, a data structure including a neural network may include any other information that determines a characteristic of a neural network. In addition, the data structure may include all types of data used or generated in the operation process of the neural network, and is not limited to the above. Computer-readable media may include computer-readable recording media and/or computer-readable transmission media. A neural network may be composed of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network is configured by including at least one or more nodes.

The data structure may include data input to the neural network. A data structure including data input to the neural network may be stored in a computer-readable medium. The data input to the neural network may include learning data input in a neural network learning process and/or input data input to the neural network in which learning is completed. Data input to the neural network may include pre-processing data and/or pre-processing target data. The preprocessing may include a data processing process for inputting data into the neural network. Accordingly, the data structure may include data to be pre-processed and data generated by pre-processing. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

The data structure may include the weights of the neural network. (In this specification, weight and parameter may be used interchangeably.) And the data structure including the weight of the neural network may be stored in a computer-readable medium. The neural network may include a plurality of weights. The weight may be variable, and may be changed by a user or an algorithm in order for the neural network to perform a desired function. For example, when one or more input nodes are interconnected to one output node by respective links, the output node sets values input to input nodes connected to the output node and links corresponding to the respective input nodes. A data value output from the output node may be determined based on the weight. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

By way of example and not limitation, the weight may include a weight variable in a neural network learning process and/or a weight in which neural network learning is completed. The variable weight in the neural network learning process may include a weight at a time point at which a learning cycle starts and/or a weight variable during the learning cycle. The weight for which neural network learning is completed may include a weight for which a learning cycle is completed. Accordingly, the data structure including the weights of the neural network may include a data structure including the weights that vary in the neural network learning process and/or the weights on which the neural network learning is completed. Therefore, it is assumed that the above-described weights and/or combinations of weights are included in the data structure including the weights of the neural network. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

The data structure including the weights of the neural network may be stored in a computer-readable storage medium (eg, memory, hard disk) after being serialized. Serialization can be the process of converting a data structure into a form that can be used by storing it on the same or different computing device and reconstructing it later. The computing device may serialize the data structure to transmit/receive data over the network. A data structure including weights of the serialized neural network may be reconstructed in the same computing device or in another computing device through deserialization. The data structure including the weight of the neural network is not limited to serialization. Furthermore, the data structure including the weights of the neural network is a data structure to increase computational efficiency while using the resources of the computing device to a minimum (e.g., B-tree, Trie, M-way search TREE, AVL tree, red-black tree). The foregoing is merely an example, and the present disclosure is not limited thereto.

The data structure may include hyper-parameters of the neural network. In addition, the data structure including the hyperparameters of the neural network may be stored in a computer-readable medium. The hyper parameter may be a variable variable by a user. Hyperparameters include, for example, a learning rate, a cost function, the number of repetitions of a learning cycle, weight initialization (eg, setting a range of weight values to be initialized for weights), hidden unit The number (eg, the number of hidden layers, the number of nodes of the hidden layer) may be included. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

One of ordinary skill in the art of this disclosure will understand that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, instructions, information, signals, bits, symbols and chips to which reference may be made in the above description include voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields particles or particles, or any combination thereof.

Those of ordinary skill in the art of the present disclosure will recognize that the various illustrative logical blocks, modules, processors, means, circuits, and algorithm steps described in connection with the embodiments disclosed herein include electronic hardware, (convenience For this purpose, it will be understood that it may be implemented by various forms of program or design code (referred to herein as software) or a combination of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. A person skilled in the art of the present disclosure may implement the described functionality in various ways for each specific application, but such implementation decisions should not be interpreted as a departure from the scope of the present disclosure.

The various embodiments presented herein may be implemented as methods, apparatus, or articles of manufacture using standard programming and/or engineering techniques. The term article of manufacture includes a computer program, carrier, or media accessible from any computer-readable storage device. For example, computer-readable storage media include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical disks (eg, CDs, DVDs, etc.), smart cards, and flash drives. memory devices (eg, EEPROMs, cards, sticks, key drives, etc.). Also, various storage media presented herein include one or more devices and/or other machine-readable media for storing information.

It is understood that the specific order or hierarchy of steps in the presented processes is an example of exemplary approaches. Based on design priorities, it is to be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure. The appended method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present disclosure is not intended to be limited to the embodiments presented herein, but is to be construed in the widest scope consistent with the principles and novel features presented herein.

Claims (12)

A method of training a model for super resolution (SR) imaging by utilizing a content density representing a proportion of content present in an image, performed by a computing device comprising at least one processor, the method comprising:
Receiving a first image and a second image having a lower resolution compared to the first image; and
matching a characteristic regarding a contents density between the first image and the second image;
including,
Matching the characteristics comprises:
outputting feature values of each of the first patch extracted from the first image and the second patch extracted from the second image; and
selecting a first patch corresponding to a characteristic related to a content density of the second patch based on each of the characteristic values;
containing,
method.
The method of claim 1,
The content of the first image and the content of the second image do not correspond to each other,
method.
delete delete The method of claim 1,
Selecting the first patch corresponding to the content density characteristic of the second patch based on the characteristic value of the first patch and the characteristic value of the second patch includes:
calculating a difference between the feature value of the first patch and the feature value of the second patch; and
selecting a first patch corresponding to a characteristic related to a content density of the second patch based on the difference in the calculated characteristic values;
containing,
method.
6. The method of claim 5,
The step of selecting the first patch corresponding to the characteristic related to the content density of the second patch based on the difference in the calculated feature values includes:
comparing the difference between the calculated feature values and a threshold value; and
selecting a first patch satisfying a predetermined condition as a patch corresponding to a content density characteristic of the second patch based on a result of the comparison;
containing,
method.
7. The method of claim 6,
The predetermined condition is
a condition for determining that a content density characteristic between the first patch and the second patch corresponds to a difference between the characteristic value of the first patch and the characteristic value of the second patch is less than the threshold value;
containing,
method.
The method of claim 1,
The feature value is
a variance value, which is one of the statistical indices indicating the characteristics of the content density of an image;
containing,
method.
The method of claim 1,
training the first model to generate a low-resolution image corresponding to the second image, based on the first image in which the second image and the content density characteristics are matched; and
training a second model to generate a high-resolution image corresponding to the first image in which the characteristic related to the content density is matched with the second image based on the output image of the first model;
further comprising,
method.
A computer program stored in a computer readable storage medium, wherein the computer program, when executed on one or more processors, utilizes a content density representing a proportion of content present in an image to train a model for super resolution (SR) imaging. to perform the following operations for performing:
receiving a first image and a second image having a lower resolution compared to the first image; and
matching a characteristic regarding a contents density between the first image and the second image;
including,
The operation of matching the characteristics is,
outputting feature values of each of the first patch extracted from the first image and the second patch extracted from the second image; and
selecting a first patch corresponding to a characteristic related to a content density of the second patch based on each of the characteristic values;
containing,
A computer program stored in a computer-readable storage medium.
A computing device that performs training of a model for super-resolution (SR) imaging by utilizing a content density that represents the proportion of content present in an image, comprising:
a processor including at least one core;
a memory containing program codes executable by the processor; and
a network unit for receiving an image;
including,
The processor is
Matching the characteristics regarding the content density between the first image and the second image of lower resolution compared to the first image,
Matching the above characteristics is:
outputting feature values of each of the first patch output from the first image and the second patch extracted from the second image; and
selecting a first patch corresponding to a characteristic related to a content density of the second patch based on each of the characteristic values;
containing,
Device.
A computer-readable recording medium storing a data structure corresponding to processed data related to a learning process of updating at least a part of a parameter of a neural network, wherein an operation of the neural network is based at least in part on the parameter, and the processing of the data Way,
Receiving a first image and a second image having a lower resolution compared to the first image; and
matching a characteristic regarding a contents density between the first image and the second image;
including,
Matching the characteristics comprises:
outputting feature values of each of the first patch extracted from the first image and the second patch extracted from the second image; and
selecting a first patch corresponding to a characteristic related to a content density of the second patch based on each of the characteristic values;
including,
wherein the content density represents the proportion of content present in the image;
A computer-readable recording medium having a data structure stored thereon.

Images