Image processing method and system and storage medium

Abstract

The invention discloses an image processing method and system and a storage medium, and the image processing method applied to a server comprises the steps: carrying out the resolution amplification and compression processing of a to-be-processed image, and obtaining a target compressed image; and issuing the target compressed image to the client, so that the client removes the compression noise of the target compressed image to obtain the target image. According to the method, the computing power resource of the server is utilized to perform super-resolution reconstruction on the to-be-processed image, and meanwhile, the occupied space of the super-resolution image is reduced through compression processing, so that the storage resource of the server is saved; therefore, the client can perform sharpening processing on the to-be-processed image on the premise of not consuming excessive computing power resources, transmission time and transmission resources, the image sharpening efficiency and the image browsing experience of a user are improved, and the method can be widely applied to the technical field of computers.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an image processing method, system and storage medium. Background Art

[0002] With the development of network transmission technology and display technology, the user experience when browsing images has been greatly enhanced. At present, the clarity of images can be improved through super-resolution reconstruction technology, but since super-resolution reconstruction requires large computing resources and the computing power of the client is limited, most super-resolution reconstruction algorithms are deployed on the server. However, after the image is processed by super-resolution reconstruction technology, while it becomes clear, its file size and the storage space occupied will also increase accordingly. When the client needs to obtain the super-resolution image from the server, it will also consume more transmission time and resources, thus affecting the efficiency of image clarity and the user's image browsing experience. Summary of the invention

[0003] The purpose of the present invention is to solve one of the technical problems existing in the prior art to at least a certain extent.

[0004] To this end, an object of embodiments of the present invention is to provide an image processing method, system and storage medium, which can improve the efficiency of image sharpening and the image browsing experience of users.

[0005] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present invention include:

[0006] In one aspect, an embodiment of the present invention provides an image processing method, comprising the following steps:

[0007] Perform resolution enlargement and compression processing on the image to be processed to obtain a target compressed image;

[0008] The target compressed image is sent to a client, so that the client removes compression noise of the target compressed image to obtain a target image.

[0009] In some embodiments of the present invention, performing resolution upscaling processing on the image to be processed is inputting the image to be processed into a pre-trained image super-resolution model for super-resolution reconstruction, and performing resolution upscaling and compression processing on the image to be processed to obtain a target compressed image includes:

[0010] Inputting the image to be processed into a pre-trained image super-resolution model for super-resolution reconstruction to obtain a target super-resolution image;

[0011] The target super-resolution image is compressed to obtain the target compressed image in a preset format.

[0012] In some embodiments of the present invention, the image super-resolution model includes a pixel inverse rearrangement module, a first convolutional layer, a cyclic feature extraction module, a feature reconstruction module, a second convolutional layer, an upsampling layer and an output layer connected in sequence, wherein the cyclic feature extraction module is used to perform cyclic feature extraction processing on the first feature map output by the first convolutional layer to obtain a second feature map with enhanced feature representation capability, and the feature reconstruction module is used to perform feature reconstruction processing on the second feature map to obtain a feature reconstructed image that amplifies image detail information.

[0013] In some embodiments of the present invention, the image super-resolution model is pre-trained by the following steps:

[0014] Acquire a plurality of preset super-resolution sample images, and perform image degradation processing on the super-resolution sample images to obtain low-resolution sample images;

[0015] Generate a training data pair according to the low-resolution sample image and the corresponding super-resolution sample image;

[0016] Inputting the training data pair into the initialized image super-resolution model to obtain a predicted super-resolution image corresponding to the low-resolution sample image;

[0017] The model parameters of the image super-resolution model are updated according to the predicted super-resolution image and the super-resolution sample image to obtain the trained image super-resolution model.

[0018] In some embodiments of the present invention, inputting the training data pair into the initialized image super-resolution model to obtain a predicted super-resolution image corresponding to the low-resolution sample image includes:

[0019] Performing image decomposition and pixel reorganization on the low-resolution sample image by the pixel inverse rearrangement module to obtain an inverse rearrangement image;

[0020] Performing image decomposition and pixel reorganization on the low-resolution sample image to obtain an inverse rearranged image;

[0021] Performing convolution processing on the inverse rearranged image to obtain a first feature map;

[0022] Performing a cyclic feature extraction process on the first feature map to obtain a second feature map;

[0023] Performing feature reconstruction processing on the second feature map to obtain a feature reconstructed image;

[0024] Performing convolution processing on the feature reconstructed image to obtain a third feature map, and then performing feature fusion processing on the first feature map and the third feature map to obtain a fourth feature map;

[0025] The fourth feature map is up-sampled to obtain the predicted super-resolution image, and the predicted super-resolution image is output.

[0026] In some embodiments of the present invention, performing cyclic feature extraction processing on the first feature map output by the first convolutional layer includes:

[0027] Performing feature extraction processing on the first feature map to obtain a first extracted feature;

[0028] The first extracted feature and the first feature Figure 1 Perform feature extraction to obtain a second extracted feature;

[0029] The first extracted features, the second extracted features and the first features are obtained Figure 1 Perform feature extraction to obtain a third extracted feature;

[0030] The obtained first extracted features, second extracted features, ..., i-th extracted features and the first feature Figure 1 Perform feature extraction from the beginning to obtain the i+1th extracted feature;

[0031] If the preset loop termination condition is not met, let i=i+1, where i is an integer greater than or equal to 3;

[0032] When the preset loop termination condition is reached, the obtained (i+1)th extracted feature is used as the second feature map.

[0033] In some embodiments of the present invention, the cyclic feature extraction module includes multiple feature extraction layers, each of the feature extraction layers includes a first convolution sublayer and a corresponding first activation function, and the cyclic feature extraction processing of the first feature map output by the first convolution layer includes:

[0034] Performing convolution processing and activation operation processing on the first feature map through the first feature extraction layer to obtain a first activation map;

[0035] Performing stacking processing on the obtained first activation map and the first feature map according to the convolution channel to obtain a first stacked feature map, and performing convolution processing and activation operation processing on the obtained first stacked feature map through the second feature extraction layer to obtain a second activation map;

[0036] The first activation map, the second activation map, and the first feature map are stacked according to the convolution channel to obtain a second stacked feature map, and the second stacked feature map is convoluted and activated by the third feature extraction layer to obtain a third activation map;

[0037] According to the convolution channel, the first activation map, the second activation map, ..., the j-th activation map and the first feature map are stacked to obtain a j-th stacked feature map, and the j-th stacked feature map is convolved and activated by the j+1-th feature extraction layer to obtain a j+1-th activation map;

[0038] In the case where the j+1th feature extraction layer is not the last feature extraction layer, let j=j+1, where j is an integer greater than or equal to 3;

[0039] When the j+1th feature extraction layer is the last feature extraction layer, the obtained j+1th activation map is used as the second feature map; or, when the j+1th feature extraction layer is the last feature extraction layer, the obtained first activation map, the second activation map, ..., the j+1th activation map and the first feature map are stacked according to the convolution channel to obtain the second feature map.

[0040] In some embodiments of the present invention, the feature reconstruction module includes a first normalization layer, a second convolution sublayer, a third convolution sublayer, a fourth convolution sublayer, a first activation sublayer, a fifth convolution sublayer, a second normalization layer, a sixth convolution sublayer, a seventh convolution sublayer, a second activation sublayer and an eighth convolution sublayer, and the feature reconstruction processing of the second feature map includes:

[0041] Performing layer normalization processing on the second feature map through the first normalization layer to obtain a first feature submap;

[0042] Through the second convolution sublayer, the third convolution sublayer and the fourth convolution sublayer, the first feature subgraph is convolved to obtain a second feature subgraph, a third feature subgraph and a fourth feature subgraph;

[0043] Multiplying the third feature sub-graph and the fourth feature sub-graph to obtain a fifth feature sub-graph, and performing activation operation processing on the fifth feature sub-graph through the first activation sub-layer to obtain a second activation sub-graph;

[0044] Multiplying the second feature subgraph and the second activation subgraph to obtain a sixth feature subgraph, and convolving the sixth feature subgraph through the fifth convolutional sublayer to obtain a seventh feature subgraph;

[0045] performing an addition process on the second feature map and the seventh feature submap to obtain an eighth feature submap, and performing a layer normalization process on the eighth feature submap through the second normalization layer to obtain a ninth feature submap;

[0046] Through the sixth convolution sublayer and the seventh convolution sublayer, convolution processing is performed on the ninth feature subgraph respectively to obtain a tenth feature subgraph and an eleventh feature subgraph;

[0047] Performing activation operation processing on the tenth feature subgraph through the second activation sublayer to obtain a twelfth feature subgraph;

[0048] The eleventh feature sub-image and the twelfth feature sub-image are multiplied to obtain a thirteenth feature sub-image, and the thirteenth feature sub-image is convolved through the eighth convolution sub-layer to obtain the feature reconstructed image.

[0049] In some embodiments of the present invention, the upsampling layer includes multiple ninth convolutional sublayers and corresponding second activation functions and a tenth convolutional sublayer, and the upsampling layer performs upsampling processing on the input fourth feature map through the following steps:

[0050] Performing transposed convolution processing and activation operation processing on the fourth feature map input to the current layer through the ninth convolution sublayer of the current layer and the second activation function to obtain the upsampled fourth feature map, and then inputting the upsampled fourth feature map to the next layer;

[0051] The tenth convolution sublayer performs transposed convolution processing on the input fourth feature map to obtain the predicted super-resolution image.

[0052] In some embodiments of the present invention, the updating of the model parameters of the image super-resolution model according to the predicted super-resolution image and the super-resolution sample image includes:

[0053] Calculating a loss value according to the predicted super-resolution image and the super-resolution sample image;

[0054] According to the loss value, the model parameters of the image super-resolution model are updated through a back-propagation algorithm.

[0055] In some embodiments of the present invention, the compressing the target super-resolution image includes:

[0056] Performing color mode conversion on the target super-resolution image to obtain a first image;

[0057] Performing chroma downsampling processing on the first image to obtain a second image;

[0058] Performing block processing on the second image to obtain a plurality of first image matrices, and then performing discrete cosine transform on the first image matrix to obtain a second image matrix;

[0059] Data quantization and Huffman coding are performed on the second image matrix to obtain the target compressed image.

[0060] On the other hand, an embodiment of the present invention provides an image processing method, applied to a client, comprising:

[0061] Uploading the image to be processed to the server, so that the server performs resolution enlargement and compression processing on the image to be processed to obtain a target compressed image;

[0062] The target compressed image sent by the server is received, and compression noise of the target compressed image is removed to obtain a target image.

[0063] In some embodiments of the present invention, removing the compression noise of the target compressed image includes:

[0064] Determine a target grayscale image according to the target compressed image;

[0065] Binarizing the target grayscale image to obtain a target binary image;

[0066] Performing a morphological closing operation on the target binary image to obtain a target noise mask image;

[0067] A noise region is determined according to the target noise mask image, and a mask is assigned to the noise region to obtain the target image.

[0068] On the other hand, an embodiment of the present invention provides an image processing system, including a client and a server, wherein:

[0069] The client is configured to upload the image to be processed to the server, and receive the target compressed image sent by the server, remove the compression noise of the target compressed image, and obtain the target image;

[0070] The server is configured to perform resolution enlargement and compression processing on the image to be processed to obtain a target compressed image, and send the target compressed image to the client.

[0071] On the other hand, an embodiment of the present invention provides an electronic device, comprising a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory, wherein the program, when executed by the processor, realizes the image processing method as described above.

[0072] On the other hand, an embodiment of the present invention further provides a storage medium, which is a computer-readable storage medium used for computer-readable storage, and the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the image processing method as described above.

[0073] The advantages and beneficial effects of the present invention will be partly given in the following description, partly become apparent from the following description, or be understood through the practice of the present invention:

[0074] In the embodiment of the present invention, a server receives an image to be processed uploaded by a client, performs resolution amplification and compression processing on the image to be processed to obtain a target compressed image, and then sends the target compressed image to the client, and removes compression noise of the target compressed image through the client to obtain the target image. On the one hand, while utilizing the computing power resources of the server to perform super-resolution reconstruction on the image to be processed, the space occupied by the super-resolution image is reduced through compression processing, which saves storage resources of the server, and also reduces the transmission time and transmission resources required for the client to obtain the super-resolution image from the server. On the other hand, by removing compression noise on the client, the clarity of the target compressed image can be restored to the same level as the initial super-resolution image of the server, and the computing power resources required for removing compression noise are significantly smaller than those for super-resolution reconstruction, so that the client can perform clarity processing on the image to be processed without consuming too much computing power resources, transmission time and transmission resources, thereby improving the efficiency of image clarity and the image browsing experience of the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solution in the embodiments of the present invention, the following introduction is made to the drawings required for use in the embodiments of the present invention. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solution of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0076] Figure 1 A flow chart of an image processing method provided by an embodiment of the present invention;

[0077] Figure 2 A flow chart of step S101 provided in an embodiment of the present invention;

[0078] Figure 3 A schematic diagram of the structure of an image super-resolution model provided by an embodiment of the present invention;

[0079] Figure 4 A flow chart of a pre-trained image super-resolution model provided in an embodiment of the present invention;

[0080] Figure 5 A flow chart of step S203 provided in an embodiment of the present invention;

[0081] Figure 6 A flow chart of performing cyclic feature extraction processing on a first feature map output by a first convolutional layer provided in an embodiment of the present invention;

[0082] Figure 7 A schematic diagram of the structure of a cyclic feature extraction module provided in an embodiment of the present invention;

[0083] Figure 8 Another flow chart of performing cyclic feature extraction processing on the first feature map output by the first convolutional layer provided in an embodiment of the present invention;

[0084] Fig. 9 A schematic diagram of the structure of a feature reconstruction module provided in an embodiment of the present invention;

[0085] Fig.10 A flow chart of performing feature reconstruction processing on a second feature map provided by an embodiment of the present invention;

[0086] Fig.11 A schematic diagram of the structure of an upsampling layer provided in an embodiment of the present invention;

[0087] Fig.12 A flow chart of upsampling the input fourth feature map provided by an embodiment of the present invention;

[0088] Fig.13 A flow chart of step S204 provided in an embodiment of the present invention;

[0089] Fig.14 A flow chart of step S1012 provided in an embodiment of the present invention;

[0090] Fig.15 Another flow chart of the image processing method provided by the embodiment of the present invention;

[0091] Fig.16 A flow chart of step S302 provided in an embodiment of the present invention;

[0092] Fig.17 A schematic diagram of the structure of an image processing system provided by an embodiment of the present invention;

[0093] Fig.18 A schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention;

[0094] Fig.19 A schematic diagram of the structure of a storage medium provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0095] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present invention. They are only examples of devices and methods consistent with some aspects of the embodiments of the present invention as detailed in the attached claims.

[0096] It is understood that the terms "first", "second", etc. used in the present invention may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0097] The terms "at least one", "multiple", "each", "any", etc. used in the present invention, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0098] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0099] Before describing the embodiments of the present invention in detail, some nouns and terms involved in the embodiments of the present invention are first described. The nouns and terms involved in the embodiments of the present invention are subject to the following explanations.

[0100] Image super-resolution reconstruction: Super-Resolution is the process of improving the resolution of the original image through hardware or software methods to obtain a high-resolution image from a low-resolution image.

[0101] JPEG: Joint Photographic Experts Group, is a product of the JPEG standard, which was developed by the International Organization for Standardization (IOS) and is a compression standard for continuous-tone still images. The JPEG format is the most commonly used image file format, with the suffix .jpg or .jpeg.

[0102] DCT: Discrete Cosine Transform, is a mathematical transformation technology widely used in signal processing and image compression. It converts a discrete signal or image into a set of cosine function coefficients, which represent the components of different frequencies in the original signal or image.

[0103] Zigzag run-length encoding: Zigzag run-length encoding is a commonly used lossless data compression technology, commonly used in image and video encoding. Its main goal is to reduce data storage space by eliminating consecutive zero values ​​and utilizing frequently occurring repeated values.

[0104] Huffman coding: Huffman coding is a commonly used lossless data compression technology that maps symbols in a data stream into variable-length codes. Huffman coding encodes symbols based on their frequency of occurrence. Symbols with higher frequencies will be assigned shorter codes, thereby achieving data compression.

[0105] Deep Learning: Deep Learning (DL) is a new research direction in the field of Machine Learning (ML). It is introduced into machine learning to make it closer to its original goal - Artificial Intelligence (AI). Deep learning is the inherent laws and representation levels of learning sample data. The information obtained in the learning process is of great help in the interpretation of data such as text, images and sounds. Its ultimate goal is to enable machines to have analytical learning capabilities like humans and to recognize data such as text, images and sounds. Deep learning has achieved many results in search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, speech, recommendation and personalization technology, and other related fields. Deep learning enables machines to imitate human activities such as seeing, hearing and thinking, solves many complex pattern recognition problems, and makes great progress in artificial intelligence related technologies.

[0106] With the development of network transmission technology and display technology, the user experience when browsing images has been greatly enhanced. At present, the clarity of images can be improved through super-resolution reconstruction technology, but since super-resolution reconstruction requires large computing resources and the computing power of the client is limited, most super-resolution reconstruction algorithms are deployed on the server. However, after the image is processed by super-resolution reconstruction technology, while it becomes clear, its file size and the storage space occupied will also increase accordingly. When the client needs to obtain the super-resolution image from the server, it will also consume more transmission time and resources, thus affecting the efficiency of image clarity and the user's image browsing experience.

[0107] In view of this, the embodiments of the present invention provide an image processing method, system, electronic device and storage medium, which receive the image to be processed uploaded by the client through the server, perform resolution enlargement and compression processing on the image to be processed to obtain a target compressed image, and then send the target compressed image to the client, and remove the compression noise of the target compressed image through the client to obtain the target image. On the one hand, while using the computing power resources of the server to perform super-resolution reconstruction on the image to be processed, the space occupied by the super-resolution image is reduced through compression processing, the storage resources of the server are saved, and the transmission time and transmission resources required for the client to obtain the super-resolution image from the server are reduced. On the other hand, by removing the compression noise on the client, the clarity of the target compressed image can be restored to the same level as the initial super-resolution image of the server, and the computing power resources required for removing the compression noise are significantly smaller than those for super-resolution reconstruction, so that the client can perform clarity processing on the image to be processed without consuming too much computing power resources, transmission time and transmission resources, thereby improving the efficiency of image clarity and the image browsing experience of the user.

[0108] The image processing method provided in the embodiment of the present invention can be applied to a terminal, or to a server, or can be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a set-top box, etc.; the server can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the image processing method, etc., but is not limited to the above forms.

[0109] The present invention can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present invention can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present invention can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0110] It should be noted that in various specific embodiments of the present invention, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with the relevant laws, regulations, and standards of the relevant countries and regions. In addition, when the embodiment of the present invention needs to obtain the user's sensitive personal information, it will obtain the user's separate permission or consent through a pop-up window or jump to a confirmation page, and after clearly obtaining the user's separate permission or consent, it will obtain the necessary user-related data for the normal operation of the embodiment of the present invention.

[0111] like Figure 1 FIG. 1 is a flow chart of an image processing method provided by an embodiment of the present invention, referring to FIG. Figure 1 The embodiment of the present invention provides an image processing method, which is applied to a server, and includes:

[0112] S101, performing resolution enlargement and compression processing on the image to be processed to obtain a target compressed image;

[0113] S102: Send the target compressed image to the client, so that the client removes compression noise of the target compressed image to obtain the target image.

[0114] Specifically, in an embodiment of the present invention, after the server performs super-resolution reconstruction on the image to be processed uploaded by the client, the super-resolution image is immediately compressed and saved on the server. The compressed and encoded target compressed image can effectively save the storage space of the server. When the client requests the server for an image after clarity processing, the compressed and encoded target compressed image is used for transmission. Since the target compressed image occupies less space than the super-resolution image before compression encoding, the required transmission time and transmission resources can be effectively reduced. Considering that image compression encoding will generate compression noise, an efficient image denoising algorithm is deployed on the client, which can effectively remove the compression noise in the target compressed image and restore the image clarity to the same level as the initial super-resolution image.

[0115] It can be recognized that the embodiment of the present invention receives the image to be processed uploaded by the client through the server, performs resolution enlargement and compression processing on the image to be processed, and obtains the target compressed image, wherein the resolution enlargement can be performed by super-resolution reconstruction of the image to be processed, and then the target compressed image is sent to the client, and the compression noise of the target compressed image is removed by the client to obtain the target image. On the one hand, while using the computing power resources of the server to reconstruct the image to be processed with super-resolution, the space occupied by the super-resolution image is reduced through compression processing, the storage resources of the server are saved, and the transmission time and transmission resources required for the client to obtain the super-resolution image from the server are reduced. On the other hand, by removing the compression noise at the client, the clarity of the target compressed image can be restored to the same level as the initial super-resolution image of the server, and the computing power resources required to remove the compression noise are significantly smaller than those for super-resolution reconstruction, so that the client can perform clarity processing on the image to be processed without consuming too much computing power resources, transmission time and transmission resources, thereby improving the efficiency of image clarity and the image browsing experience of the user.

[0116] like Figure 2 FIG. 1 is a flow chart of step S101 provided in an embodiment of the present invention, referring to FIG. Figure 2 In some embodiments of the present invention, performing resolution upscaling processing on the image to be processed is inputting the image to be processed into a pre-trained image super-resolution model for super-resolution reconstruction, performing resolution upscaling and compression processing on the image to be processed to obtain a target compressed image, including:

[0117] S1011, inputting the image to be processed into a pre-trained image super-resolution model for super-resolution reconstruction to obtain a target super-resolution image;

[0118] S1012: compress the target super-resolution image to obtain a target compressed image in a preset format.

[0119] Specifically, the images to be processed uploaded by the client to the server are mostly obtained through the client's APP or through image capture, image screenshots, etc. The APP usually compresses the image resolution to a size suitable for the client's display screen before transmitting it for display. Image capture depends on the resolution of the client's camera, and the resolution of the image screenshot is directly equivalent to the resolution of the client's display screen.

[0120] It can be seen that the images obtained by the above three methods may have too low resolution. Therefore, in the embodiment of the present invention, the client uploads the image to be processed to the server, and the image super-resolution model deployed on the server performs super-resolution reconstruction on the image to be processed, thereby enlarging the resolution of the image to be processed and obtaining the target super-resolution image. The image super-resolution model of the embodiment of the present invention is implemented by deep learning. The model structure and training process of the image super-resolution model of the embodiment of the present invention are described below.

[0121] like Figure 3 FIG. 1 is a schematic diagram of the structure of an image super-resolution model provided by an embodiment of the present invention, referring to FIG. Figure 3 In some embodiments of the present invention, the image super-resolution model includes a pixel inverse rearrangement module, a first convolutional layer, a cyclic feature extraction module, a feature reconstruction module, a second convolutional layer, an upsampling layer and an output layer connected in sequence, wherein the cyclic feature extraction module is used to perform cyclic feature extraction processing on the first feature map output by the first convolutional layer to obtain a second feature map with enhanced feature representation capability, and the feature reconstruction module is used to perform feature reconstruction processing on the second feature map to obtain a feature reconstructed image with amplified image detail information.

[0122] Specifically, the image super-resolution model of the embodiment of the present invention is obtained by training with a deep neural network. The specific architecture of the model is as follows: Figure 3 As shown, the pixel inverse rearrangement module is used to reduce the image resolution to facilitate the extraction of image features, the cyclic feature extraction module is used to cyclically extract image features to enhance the feature representation capability, the feature reconstruction module is used to reconstruct the image according to the extracted image features to amplify the image detail information, and the upsampling layer is used to upsample the reconstructed image to generate a high-resolution image.

[0123] It should be noted that the cyclic feature extraction module gradually extracts image features by cyclically performing multiple feature extractions, so that the extracted image features are more representative and distinguishable, thereby obtaining a second feature map with enhanced feature representation capabilities.

[0124] like Figure 4 FIG. 1 is a flow chart of a pre-trained image super-resolution model provided by an embodiment of the present invention, referring to FIG. Figure 4 In some embodiments of the present invention, the image super-resolution model is pre-trained by the following steps:

[0125] S201, obtaining a plurality of preset super-resolution sample images, and performing image degradation processing on the super-resolution sample images to obtain low-resolution sample images;

[0126] S202, generating a training data pair according to a low-resolution sample image and a corresponding super-resolution sample image;

[0127] S203, inputting the training data pair into the initialized image super-resolution model to obtain a predicted super-resolution image corresponding to the low-resolution sample image;

[0128] S204: Update the model parameters of the image super-resolution model according to the predicted super-resolution image and the super-resolution sample image to obtain a trained image super-resolution model.

[0129] Specifically, a plurality of preset super-resolution sample images are obtained, and the resolution of the super-resolution sample images is reduced by a degradation algorithm, thereby simulating the initial low-resolution image obtained by the client. Considering that there are various reasons for image quality degradation in the real world, the training data generated by using a specific image degradation method often does not work well when processing real pictures. In order to handle the super-resolution task of low-resolution images caused by various reasons and improve the generalization ability of the image super-resolution model, the embodiment of the present invention uses different image degradation methods to simulate different low-resolution images to train the image super-resolution model.

[0130] In some optional embodiments, the degradation algorithm used in the embodiments of the present invention includes but is not limited to blur operation, scaling operation, superimposed noise operation and compression operation, which are as follows:

[0131] Fuzzy operation: According to the random selection method, at least one of Gaussian blur and Sinc filter blur is selected for operation;

[0132] Scaling operation: according to the random selection method, at least one of bilinear interpolation, bicubic interpolation and regional interpolation is selected for operation;

[0133] Superposition noise operation: According to the random selection method, at least one of Gaussian noise and Poisson noise is selected for operation;

[0134] Image compression operation: compress the image, and the compression factor can be set freely.

[0135] It should be noted that the above-mentioned blurring operation, scaling operation, noise superposition operation and compression operation can be used alone for the image degradation processing in the embodiment of the present invention, or can be arbitrarily combined for the image degradation processing in the embodiment of the present invention, and the embodiment of the present invention is not limited to this.

[0136] Based on the low-resolution sample images obtained through image degradation and the corresponding super-resolution sample images, training data pairs required for training the image super-resolution model can be generated; the training data pairs are input into the initialized image super-resolution model to obtain the predicted super-resolution images corresponding to the low-resolution sample images; the model parameters of the image super-resolution model are updated according to the predicted super-resolution images and the super-resolution sample images to obtain the trained image super-resolution model.

[0137] like Figure 5 FIG. 1 is a flow chart of step S203 provided in an embodiment of the present invention, referring to FIG. Figure 5 In some embodiments of the present invention, the training data pair is input into the initialized image super-resolution model to obtain a predicted super-resolution image corresponding to the low-resolution sample image, including:

[0138] S2031, performing image decomposition and pixel reorganization on the low-resolution sample image to obtain an inverse rearrangement image;

[0139] S2032, performing convolution processing on the inverse rearrangement image to obtain a first feature map;

[0140] S2033, performing a cyclic feature extraction process on the first feature map to obtain a second feature map;

[0141] S2034, performing feature reconstruction processing on the second feature map to obtain a feature reconstructed image;

[0142] S2035, performing convolution processing on the feature reconstructed image to obtain a third feature map, and then performing feature fusion processing on the first feature map and the third feature map to obtain a fourth feature map;

[0143] S2036. Upsample the fourth feature map to obtain a predicted super-resolution image, and output the predicted super-resolution image.

[0144] Specifically, the image super-resolution model first decomposes and reorganizes the input low-resolution sample image through the pixel inverse rearrangement module to reduce the image resolution. The main benefits of this are as follows:

[0145] 1) Reduce computational complexity: Reducing the dimensionality of high-resolution image space helps reduce computational complexity in the feature extraction process and helps improve the computational efficiency of the network.

[0146] 2) More compact feature representation: By concentrating spatial information into the number of channels, a more compact feature representation can be obtained while maintaining the original information, which helps to improve the storage and computational efficiency of feature representation.

[0147] 3) Improve feature combination: Encoding spatial information into the number of channels can increase feature combination in high-dimensional feature space, which helps to improve the expressiveness and representation ability of features.

[0148] Then, the number of feature map channels of the model is further increased through the first convolution layer, and the input inverse rearrangement image is further subjected to feature extraction to obtain a first feature map; the first feature map is subjected to cyclic feature extraction through a cyclic feature extraction module to obtain a second feature map; the second feature map is subjected to feature reconstruction processing through a feature reconstruction module to obtain a feature reconstructed image; the feature reconstructed image is subjected to convolution processing through a second convolution layer to obtain a third feature map, and then the first feature map output by the first convolution layer and the third feature map output by the second convolution layer are subjected to feature fusion processing, such as image addition, to obtain a fourth feature map.

[0149] It should be noted that by performing feature reconstruction processing on the second feature map extracted by the cyclic feature extraction module through the feature reconstruction module, image noise can be suppressed and detail information can be amplified, so that when the first feature map output by the first convolutional layer and the third feature map output by the second convolutional layer are subsequently subjected to feature fusion processing, image noise can be effectively removed and the lost detail information of the image can be restored.

[0150] Finally, the fourth feature map is upsampled through the upsampling layer to obtain a predicted super-resolution image, and the predicted super-resolution image is output through the output layer.

[0151] like Figure 6 FIG. 1 is a flow chart of performing cyclic feature extraction processing on the first feature map output by the first convolutional layer according to an embodiment of the present invention, referring to FIG. Figure 6 , performing cyclic feature extraction processing on the first feature map output by the first convolutional layer, including:

[0152] S20331, performing feature extraction processing on the first feature map to obtain a first extracted feature;

[0153] S20332, the obtained first extracted feature and the first feature Figure 1 Perform feature extraction to obtain a second extracted feature;

[0154] S20333, the obtained first extracted features, the second extracted features and the first features Figure 1 Perform feature extraction to obtain a third extracted feature;

[0155] S20334, the obtained first extracted feature, the second extracted feature, ..., the i-th extracted feature and the first feature Figure 1Feature extraction is performed from the beginning to obtain the i+1th extracted feature; when the preset loop termination condition is not met, let i=i+1, where i is an integer greater than or equal to 3; when the preset loop termination condition is met, the obtained i+1th extracted feature is used as the second feature map.

[0156] Specifically, feature extraction is performed on the first feature map to obtain a first extracted feature, the first extracted feature and the first feature map are stacked and feature extracted together to obtain a second extracted feature, and then the first extracted feature, the second extracted feature and the first feature map are stacked and feature extracted together again to obtain a third extracted feature. And so on, each subsequent round extracts the first extracted feature, the second extracted feature, ..., the i-th extracted feature and the first feature map. Figure 1 Feature extraction is performed from the beginning to the end to obtain the i+1th extracted feature. If the preset loop termination condition is not reached (such as the preset number of feature extraction times is reached, or the preset number of feature extraction layers have completed feature extraction), set i=i+1 and perform the next round of feature extraction. If the preset loop termination condition is reached, the i+1th extracted feature obtained in the last round is used as the second feature map.

[0157] Further, in some embodiments of the present invention, the first extracted feature may be a first activation map; the first activation map and the first feature Figure 1 The first activation map and the first feature map are stacked to obtain a first stacked feature map, and the first stacked feature map is extracted to obtain a second activation map. The first feature map output by the first convolutional layer is subjected to cyclic feature extraction, including:

[0158] Performing feature extraction on the first feature map to obtain a first activation map;

[0159] The obtained first activation map and the first feature map are stacked to obtain a first stacked feature map, and feature extraction is performed on the obtained first stacked feature map to obtain a second activation map;

[0160] The obtained first activation map, the second activation map, and the first feature map are stacked to obtain a second stacked feature map, and feature extraction is performed on the obtained second stacked feature map to obtain a third activation map;

[0161] The obtained first activation map, the second activation map, ..., the j-th activation map and the first feature map are stacked to obtain a j-th stacked feature map, and feature extraction is performed on the obtained j-th stacked feature map to obtain a j+1-th activation map;

[0162] If the preset loop termination condition is not met, let j=j+1, where j is an integer greater than or equal to 3;

[0163] When a preset loop termination condition is reached, the obtained j+1th activation map is used as the second feature map; or, the obtained first activation map, the second activation map, ..., the j+1th activation map and the first feature map are stacked to obtain the second feature map.

[0164] Specifically, feature extraction is performed on the first feature map to obtain a first activation map, the first activation submap and the first feature map are stacked to obtain a first stacked feature map, feature extraction is performed on the first stacked feature map to obtain a second activation map, and then the first activation map, the second activation map and the first feature map are stacked and feature extracted again to obtain a third activation map. By analogy, each subsequent round stacks the obtained first activation map, the second activation map, ..., the jth activation map and the first feature map to obtain the jth stacked feature map, and then performs feature extraction on the obtained jth stacked feature map to obtain the j+1th activation map; if the preset loop termination condition is not reached (such as reaching the preset number of feature extraction times, or the preset number of feature extraction layers have completed feature extraction), let j=j+1, and perform the next round of feature extraction; if the preset loop termination condition is reached, the j+1th activation map obtained in the last round is used as the second feature map, or the obtained first activation map, the second activation map, ..., the j+1th activation map and the first feature map are stacked to obtain the second feature map.

[0165] like Figure 7 FIG. 4 is a schematic diagram of the structure of a cyclic feature extraction module provided in an embodiment of the present invention. Figure 8 FIG. 1 is another flowchart of performing cyclic feature extraction processing on the first feature map output by the first convolutional layer according to an embodiment of the present invention, referring to FIG. Figure 7 and 8 In some embodiments of the present invention, the cyclic feature extraction module includes a plurality of feature extraction layers, each of which includes a first convolution sublayer and a corresponding first activation function, and performs cyclic feature extraction processing on the input first feature map, including:

[0166] S20335, performing convolution processing and activation operation processing on the first feature map through the first feature extraction layer to obtain a first activation map;

[0167] S20336, stacking the obtained first activation map and the first feature map according to the convolution channel to obtain a first stacked feature map, and performing convolution processing and activation operation processing on the obtained first stacked feature map through the second feature extraction layer to obtain a second activation map;

[0168] S20337, stacking the first activation map, the second activation map, and the first feature map obtained according to the convolution channel to obtain a second stacked feature map, and performing convolution processing and activation operation processing on the second stacked feature map obtained through a third feature extraction layer to obtain a third activation map;

[0169] S20338. Perform stacking processing on the obtained first activation map, the second activation map, ..., the j-th activation map, and the first feature map according to the convolution channel to obtain the j-th stacked feature map, and perform convolution processing and activation operation processing on the obtained j-th stacked feature map through the j+1-th feature extraction layer to obtain the j+1-th activation map; when the j+1-th feature extraction layer is not the last feature extraction layer, let j=j+1, where j is an integer greater than or equal to 3; when the j+1-th feature extraction layer is the last feature extraction layer, use the obtained j+1-th activation map as the second feature map, or, when the j+1-th feature extraction layer is the last feature extraction layer, perform stacking processing on the obtained first activation map, the second activation map, ..., the j+1-th activation map, and the first feature map according to the convolution channel to obtain the second feature map.

[0170] Specifically, Figure 7 As shown, the cyclic feature extraction module of the embodiment of the present invention includes multiple feature extraction layers, each of which includes a convolution sublayer and a corresponding activation function, wherein the convolution sublayer and the activation function of each feature extraction layer may be the same or different. The convolution sublayer of each feature extraction layer performs convolution processing on the feature map input to the layer, and then performs activation operation through the activation function to obtain the corresponding activation map, and then stacks the activation map of the layer with the activation maps output by all previous feature extraction layers and the first feature map according to the convolution channel to obtain a stacked feature map, and then inputs the stacked feature map to the next layer for feature extraction, until the last layer of the cyclic feature extraction module outputs the stacked feature map, and then performs feature extraction on the obtained stacked feature map to obtain the activation map as the second feature map, or, the stacked feature map output by the last layer is used as the cyclic feature extraction module to extract the second feature map.

[0171] It can be understood that the cyclic feature extraction module of an embodiment of the present invention completes the convolution processing and activation operation process respectively through the convolution sublayer and activation function of the feature extraction layer, thereby obtaining the activation map corresponding to the current feature extraction layer, and stacking the activation map of the current feature extraction layer and the activation maps output by all previous feature extraction layers and the first feature map according to the convolution channel to obtain the stacked feature map corresponding to the current feature extraction layer. When the current feature extraction layer is not the last feature extraction layer, the stacked feature map corresponding to the current feature extraction layer continues to be input to the next feature extraction layer for feature extraction and stacking processing. When the current feature extraction layer is the last feature extraction layer, the stacked feature map corresponding to the current feature extraction layer is output as the second feature map. It should be noted that, unlike the aforementioned embodiment in which the extracted features / activation map of the last round of cyclic feature extraction is used as the second feature map, after the corresponding activation map is obtained in the last feature extraction layer, the embodiment of the present invention can still be stacked with the activation maps output by all previous feature extraction layers and the first feature map, and the stacked feature map finally obtained is used as the second feature map, which can further enhance the feature representation capability of the second feature map.

[0172] In some embodiments of the present invention, the first activation function may adopt an LReLU activation function.

[0173] It should be noted that the number of first convolution sublayers and the convolution kernel parameters of each first convolution sublayer in the cyclic feature extraction module can be set according to actual needs, and the embodiment of the present invention does not limit this. In some optional embodiments, 64 first convolution sublayers are set to perform 64 cyclic feature extractions to obtain a second feature map.

[0174] like Fig. 9 FIG. 4 is a schematic diagram of the structure of a feature reconstruction module provided in an embodiment of the present invention. Fig.10 FIG. 1 is a flow chart of a feature reconstruction process for a second feature map provided by an embodiment of the present invention, referring to FIG. Fig. 9 and 10 In some embodiments of the present invention, the feature reconstruction module includes a first normalization layer, a second convolution sublayer, a third convolution sublayer, a fourth convolution sublayer, a first activation sublayer, a fifth convolution sublayer, a second normalization layer, a sixth convolution sublayer, a seventh convolution sublayer, a second activation sublayer and an eighth convolution sublayer, and performs feature reconstruction processing on the second feature map, including:

[0175] S20341, performing layer normalization processing on the second feature map through a first normalization layer to obtain a first feature sub-map;

[0176] S20342, performing convolution processing on the first feature subgraph through the second convolution sublayer, the third convolution sublayer, and the fourth convolution sublayer, respectively, to obtain a second feature subgraph, a third feature subgraph, and a fourth feature subgraph;

[0177] S20343, multiplying the third feature sub-graph and the fourth feature sub-graph to obtain a fifth feature sub-graph, and performing activation operation processing on the fifth feature sub-graph through the first activation sub-layer to obtain a second activation sub-graph;

[0178] S20344, multiplying the second feature sub-graph and the second activation sub-graph to obtain a sixth feature sub-graph, and convolving the sixth feature sub-graph through a fifth convolutional sub-layer to obtain a seventh feature sub-graph;

[0179] S20345, adding the second feature map and the seventh feature sub-map to obtain an eighth feature sub-map, and performing layer normalization processing on the eighth feature sub-map through a second normalization layer to obtain a ninth feature sub-map;

[0180] S20346, performing convolution processing on the ninth feature subgraph through the sixth convolution sublayer and the seventh convolution sublayer, respectively, to obtain a tenth feature subgraph and an eleventh feature subgraph;

[0181] S20347, performing activation operation processing on the tenth feature subgraph through the second activation sublayer to obtain a twelfth feature subgraph;

[0182] S20348. Multiply the eleventh feature sub-image and the twelfth feature sub-image to obtain a thirteenth feature sub-image, and perform convolution processing on the thirteenth feature sub-image through the eighth convolution sub-layer to obtain a feature reconstructed image.

[0183] In some embodiments of the present invention, both the first normalization layer and the second normalization layer can use LayerNorm, which is a common regularization technology in deep learning models. It solves the gradient vanishing problem by normalizing the output of neurons, accelerates training, and has a significant effect on stabilizing training and improving model performance. The calculation process of LayerNorm includes calculating the mean, variance, standardization, and affine transformation. The specific calculation process is not described in detail in the embodiments of the present invention.

[0184] The second convolution sublayer, the third convolution sublayer, the fourth convolution sublayer, the sixth convolution sublayer and the seventh convolution sublayer of the embodiment of the present invention are all provided with 1*1 and 3*3 convolution kernels, which are used to perform a 1*1 convolution and a 3*3 convolution on the input image. The fifth convolution sublayer and the eighth convolution sublayer of the embodiment of the present invention are both provided with a 1*1 convolution kernel, which are used to perform a 1*1 convolution on the input image.

[0185] In one embodiment of the present invention, the first activation sublayer may adopt a SoftMax activation function to perform an activation operation on a fifth feature subgraph obtained by multiplying the third feature subgraph and the fourth feature subgraph to obtain a second activation subgraph; the second activation sublayer may adopt a GELU activation function to perform an activation operation on the tenth feature subgraph output by the sixth convolution sublayer to obtain a twelfth feature subgraph.

[0186] like Fig.11 FIG. 4 is a schematic diagram of the structure of the upsampling layer provided in an embodiment of the present invention. Fig.12 FIG. 1 is a flow chart of performing upsampling processing on the fourth feature map of the input according to an embodiment of the present invention, referring to FIG. Fig.11 and 12 In some embodiments of the present invention, the upsampling layer includes multiple ninth convolutional sublayers and corresponding second activation functions and a tenth convolutional sublayer, and the upsampling layer performs upsampling processing on the input fourth feature map through the following steps:

[0187] S20361, performing transposed convolution processing and activation operation processing on the fourth feature map input to the current layer through the ninth convolution sublayer and the second activation function of the current layer to obtain an upsampled fourth feature map, and then input the upsampled fourth feature map to the next layer;

[0188] S20362. Perform transposed convolution processing on the input fourth feature map through the tenth convolution sublayer to obtain a predicted super-resolution image.

[0189] Specifically, Fig.11 As shown, the upsampling layer of the embodiment of the present invention includes multiple ninth convolution sublayers and one tenth convolution sublayer, and the ninth convolution sublayer and the tenth convolution sublayer are both used to perform transposed convolution (deconvolution) processing on the input image, thereby achieving image upsampling. Each ninth convolution sublayer has a corresponding second activation function. Each ninth convolution sublayer performs a transposed convolution process on the fourth feature map input to the layer, and then performs an activation operation through the second activation function to obtain the fourth feature map after upsampling of the layer, and then inputs it to the next layer to continue the transposed convolution process and activation operation until the next layer is the tenth convolution sublayer; the fourth feature map input to the layer is transposed convolution processed by the tenth convolution sublayer, and a predicted super-resolution image can be obtained.

[0190] In some embodiments of the present invention, the second activation function may adopt an LReLU activation function.

[0191] It should be noted that the number of the ninth convolution sublayer in the upsampling layer can be set according to actual needs, and the embodiment of the present invention is not limited to this.

[0192] like Fig.13FIG. 1 is a flow chart of step S204 provided in an embodiment of the present invention, referring to FIG. Fig.13 In some embodiments of the present invention, updating the model parameters of the image super-resolution model according to the predicted super-resolution image and the super-resolution sample image includes:

[0193] S2041, calculating a loss value according to the predicted super-resolution image and the super-resolution sample image;

[0194] S2042. Update the model parameters of the image super-resolution model through a back-propagation algorithm according to the loss value.

[0195] Specifically, the training loss value is calculated based on the predicted super-resolution image and the super-resolution sample image in the training data pair. In some embodiments of the invention, one loss function or a combination of multiple loss functions can be selected from L1 loss, VGG loss, GAN loss and MS-SSIM loss to determine the training loss value. Based on the training loss value, the parameters of the model are updated using a back propagation algorithm. A trained image super-resolution model can be obtained after several rounds of iterations. The specific number of iterations can be set in advance, or the training is considered to be completed when the loss value reaches a preset threshold and the test set meets the accuracy requirements.

[0196] The image to be processed is input into the trained image super-resolution model for super-resolution reconstruction to obtain the target super-resolution image.

[0197] like Fig.14 FIG. 1 is a flow chart of step S1012 provided in an embodiment of the present invention, referring to FIG. Fig.14 In some embodiments of the present invention, the target super-resolution image is compressed, including:

[0198] S10121, performing color mode conversion on the target super-resolution image to obtain a first image;

[0199] S10122, performing chroma downsampling processing on the first image to obtain a second image;

[0200] S10123, performing block processing on the second image to obtain a plurality of first image matrices, and then performing discrete cosine transform on the first image matrices to obtain a second image matrix;

[0201] S10124. Perform data quantization and Huffman coding on the second image matrix to obtain a target compressed image.

[0202] Specifically, common image compression algorithms include JPEG compression, WEBP compression, AVIF compression, and intelligent compression algorithms based on neural networks. It should be noted that the lossless compression technology of images has limited compression on images, and the storage space and network transmission time saved are not much, while the lossy compression algorithm of images can greatly reduce the storage space of images, and the required transmission time and transmission resources will also be reduced accordingly, but noise will be generated, so the compression noise needs to be removed on the client. Considering that after the target super-resolution image is compressed and encoded on the server, the compression noise needs to be removed on the client to avoid affecting the user's browsing experience, when designing the compression encoding algorithm on the server, it is also necessary to take into account the computing power resources that the client can provide to the denoising algorithm.

[0203] Currently, the more popular denoising algorithms are those based on deep learning, such as image quality restoration algorithms based on convolutional neural networks and denoising methods based on generative adversarial networks. Although they can achieve good results, the computing power resources of the client are limited. Some denoising algorithms based on deep learning cannot be run on the terminal, and some deep learning algorithms that can be run on the client also take a long time to calculate, which also affects the user's image browsing experience.

[0204] In some embodiments of the present invention, the generated target super-resolution image is compressed and encoded in JPEG format on the server side to obtain a target compressed image. In this way, the client side only needs to remove the compression noise generated by the JPEG compression encoding to make the restored target image reach the same pixel level as the target super-resolution image. Therefore, the client side only needs to deploy a denoising algorithm to remove the compression noise generated by the JPEG compression encoding.

[0205] First, the target super-resolution image is converted into a color mode, and the RGB color mode is converted into a YCbCr color mode to obtain a first image. R, G, and B in the RGB color mode represent the three components of red, green, and blue, respectively. Y in the YCbCr color mode represents brightness, and Cb and Cr represent saturation and chroma, respectively. A key principle of lossy compression technology is to process important information and unimportant information separately. For the human eye, the light and dark changes in the image are easier to be perceived than the color changes. Therefore, the embodiment of the present invention converts the RGB color mode into the YCbCr color mode to facilitate subsequent compression encoding in JPEG format.

[0206] Secondly, the first image is subjected to chroma downsampling to obtain a second image. Since the human eye is much more sensitive to brightness changes than to color changes, the Y component is more important than the Cb and Cr components in the compression coding process. The embodiment of the present invention can adopt two sampling methods: YUV411 and YUV422, which represent that the data sampling ratios of the three components Y, Cb, and Cr are 4:1:1 and 4:2:2 respectively. Although part of the chroma and saturation information is lost through chroma downsampling, the brightness information that the human eye is most sensitive to is retained, and the amount of data storage and subsequent calculations are also reduced.

[0207] Then, the second image is processed in blocks to obtain multiple first image matrices, and then the first image matrix is ​​subjected to discrete cosine transform to obtain the second image matrix. Block processing is to divide the second image into multiple 8*8 pixel matrices to facilitate subsequent discrete cosine transform. Discrete cosine transform (DCT) is a transform coding method commonly used in bit rate compression. The Fourier transform of any continuous real symmetric function contains only cosine terms. Therefore, the cosine transform has a clear physical meaning like the Fourier transform. The encoding process of JPEG requires a forward discrete cosine transform, while the decoding process requires an inverse discrete cosine transform. In an embodiment of the present invention, the multiple first image matrices obtained by the block processing are respectively subjected to discrete cosine transform to obtain multiple second image matrices.

[0208] Finally, the second image matrix is ​​subjected to data quantization and Huffman coding to obtain the target compressed image. After the second image matrix is ​​obtained by discrete cosine transform, a quantization stage is required before the coding process can be entered. The quantization stage requires two 8*8 quantization matrix data, one is a frequency coefficient specifically for processing brightness, and the other is a frequency coefficient for chrominance. The frequency coefficient is divided by the value of the quantization matrix and then rounded to complete the quantization process. After the frequency coefficient is quantized, the frequency coefficient is converted from a floating point number to an integer, which is convenient for performing the final coding. It is not difficult to find that after the quantization stage, all data only retain integer approximations, and some data content is lost again. In the JPEG algorithm, due to different accuracy requirements for brightness and chrominance, different quantization tables can be used for brightness and chrominance respectively, the former is fine quantization, and the latter is coarse quantization. The last step of JPEG compression is to perform Huffman coding on the data, and adjust the coding length of the element according to the frequency of use of the element in the data to obtain a higher compression ratio. The embodiment of the present invention effectively reduces the occupied space of the target super-resolution image through JPEG compression coding.

[0209] In some optional embodiments, Zigzag run-length encoding may be performed before Huffman encoding to reduce data storage space by eliminating consecutive zero values ​​and utilizing frequently occurring repeated values, which will not be described in detail in the embodiments of the present invention.

[0210] The above is a description of the steps of the image processing method applied to the server. It can be understood that the embodiment of the present invention receives the image to be processed uploaded by the client through the server, performs resolution amplification and compression processing on the image to be processed, obtains the target compressed image, and then sends the target compressed image to the client, removes the compression noise of the target compressed image through the client, and obtains the target image. On the one hand, while using the computing power resources of the server to perform super-resolution reconstruction on the image to be processed, the space occupied by the super-resolution image is reduced through compression processing, the storage resources of the server are saved, and the transmission time and transmission resources required for the client to obtain the super-resolution image from the server are reduced. On the other hand, by removing the compression noise at the client, the clarity of the target compressed image can be restored to the same level as the initial super-resolution image of the server, and the computing power resources required to remove the compression noise are significantly smaller than those for super-resolution reconstruction, so that the client can perform clarity processing on the image to be processed without consuming too much computing power resources, transmission time and transmission resources, thereby improving the efficiency of image clarity and the image browsing experience of the user.

[0211] Corresponding to the above-mentioned image processing method applied to the server, the embodiment of the present invention also provides an image processing method applied to the client. Fig.15 FIG. 1 is another flow chart of an image processing method provided by an embodiment of the present invention. Fig.15 The embodiment of the present invention provides an image processing method, which is applied to a client and includes:

[0212] S301, uploading the image to be processed to the server, so that the server performs resolution enlargement and compression processing on the image to be processed to obtain a target compressed image;

[0213] S302: Receive a target compressed image sent by the server, remove compression noise from the target compressed image, and obtain a target image.

[0214] Specifically, in an embodiment of the present invention, a client uploads an image to be processed to a server, and after the server performs super-resolution reconstruction on the image to be processed, the super-resolution image is immediately compressed and saved on the server. The compressed and encoded target compressed image can effectively save storage space on the server. When the client requests the server for an image after clarity processing, the compressed and encoded target compressed image is used for transmission. Since the target compressed image occupies less space than the super-resolution image before compression encoding, the required transmission time and transmission resources can be effectively reduced. Considering that image compression encoding will generate compression noise, an efficient image denoising algorithm is deployed on the client, which can effectively remove the compression noise in the target compressed image and restore the image clarity to the same level as the initial super-resolution image.

[0215] like Fig.16 FIG. 1 is a flow chart of step S302 provided in an embodiment of the present invention, referring to FIG. Fig.16 In some embodiments of the present invention, removing compression noise of a target compressed image includes:

[0216] S3021, determining a target grayscale image according to the target compressed image;

[0217] S3022, binarizing the target grayscale image to obtain a target binary image;

[0218] S3023, performing a morphological closing operation on the target binary image to obtain a target noise mask image;

[0219] S3024. Determine the noise area according to the target noise mask image, and assign a mask value to the noise area to obtain the target image.

[0220] In some embodiments of the present invention, the target compressed image is obtained through processes such as color mode conversion, chroma downsampling, discrete cosine transform, and Huffman coding, and compression noise inevitably exists, so denoising processing is required on the client side.

[0221] First, a target grayscale image is determined according to the target compressed image, and the target grayscale image is binarized to obtain a target binary image. If the target compressed image itself is a grayscale image, the target compressed image is used as the target grayscale image. If the target compressed image is not a grayscale image, the target compressed image can be converted into a target grayscale image through channel conversion processing, and all grayscale levels are retained. Binarization processing can convert pixels in a grayscale image into only black and white colors, highlighting the boundaries and contours in the image. In an embodiment of the present invention, the noise in the target compressed image can be made more obvious through binarization processing.

[0222] Then, the target binary image is subjected to morphological closing operation to obtain the target noise mask image. Morphological closing operation is the process of image dilation and erosion in sequence. Through morphological closing operation, small holes in the binary image can be removed, and disconnected parts in the image can be connected to avoid noise omission.

[0223] Finally, the noise area is determined according to the target noise mask image, and the mask is assigned to the noise area to obtain the target image. The noise area can be quickly extracted from the target noise mask image, and all the extracted noise areas are assigned to pixel points with specific pixel values ​​to obtain the target image.

[0224] It should be noted that, for a target compressed image whose initial state is a grayscale image, there is no need to perform channel conversion processing. The target image can be obtained by subsequently assigning white (255) pixels to the extracted noise area in the target noise mask image. For a target compressed image whose initial state is not a grayscale image, after obtaining the target noise mask image, it is necessary to determine the noise area in the initial target compressed image based on the target noise mask image (specifically, noise extraction can be performed in the target noise mask image first, and then the corresponding area in the initial target compressed image is used as the noise area). Subsequently, a mask is assigned to the noise area in the target compressed image. When assigning the mask, the specific pixel value assigned is determined by calculating the average pixel value of the neighborhood pixels of each noise point, and the target image can be obtained.

[0225] It can be recognized that the embodiment of the present invention uploads the image to be processed to the server, so that the server performs resolution amplification and compression processing on the image to be processed to obtain a target compressed image, and then receives the target compressed image sent by the server, removes the compression noise of the target compressed image, and obtains the target image. On the one hand, while using the computing power resources of the server to perform super-resolution reconstruction on the image to be processed, the space occupied by the super-resolution image is reduced through compression processing, which saves the storage resources of the server and reduces the transmission time and transmission resources required for the client to obtain the super-resolution image from the server. On the other hand, by removing the compression noise on the client, the clarity of the target compressed image can be restored to the same level as the initial super-resolution image of the server, and the computing power resources required for removing the compression noise are significantly smaller than those for super-resolution reconstruction, so that the client can perform clarity processing on the image to be processed without consuming too much computing power resources, transmission time and transmission resources, thereby improving the efficiency of image clarity and the user's image browsing experience.

[0226] like Fig.17 FIG. 1 is a schematic diagram of the structure of an image processing system provided by an embodiment of the present invention, referring to FIG. Fig.17 , an embodiment of the present invention provides an image processing system, including a client and a server, wherein:

[0227] The client is configured to upload the image to be processed to the server, and receive the target compressed image sent by the server, remove the compression noise of the target compressed image, and obtain the target image;

[0228] The server is configured to perform resolution enlargement and compression processing on the image to be processed to obtain a target compressed image, and send the target compressed image to the client.

[0229] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0230] The embodiment of the present invention further provides an electronic device, the electronic device comprising: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory, and the program is executed by the processor to realize the above-mentioned image processing method. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0231] like Fig.18 FIG. 1 is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention, referring to FIG. Fig.18 , an embodiment of the present invention provides an electronic device, including:

[0232] The processor 1801 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention;

[0233] The memory 1802 may be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1802 may store an operating system and other application programs. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1802, and is called by the processor 1801 to execute the image processing method of the embodiment of the present invention;

[0234] Input / output interface 1803, used to implement information input and output;

[0235] The communication interface 1804 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0236] A bus 1805 that transmits information between the various components of the device (e.g., the processor 1801, the memory 1802, the input / output interface 1803, and the communication interface 1804);

[0237] The processor 1801 , the memory 1802 , the input / output interface 1803 and the communication interface 1804 are connected to each other in communication within the device via the bus 1805 .

[0238] like Fig.19 FIG. 1 is a schematic diagram of a storage medium according to an embodiment of the present invention. Fig.19 The embodiment of the present invention further provides a storage medium, which is a computer-readable storage medium used for computer-readable storage. The storage medium stores one or more programs 1901, and the one or more programs 1901 can be executed by one or more processors to implement the above-mentioned image processing method.

[0239] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0240] The embodiment of the present invention further provides a computer program product or a computer program, wherein the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes Figure 1 Or the method shown in 15.

[0241] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the above-mentioned boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0242] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified to the contrary, one or more of the above-mentioned functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0243] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the above methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0244] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0245] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the above-mentioned program is printed, since the above-mentioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or processing in other suitable ways as necessary, and then stored in a computer memory.

[0246] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0247] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0248] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0249] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. An image processing method, characterized in that: Applied to the server, including: Perform resolution enlargement and compression processing on the image to be processed to obtain a target compressed image; The target compressed image is sent to a client, so that the client removes compression noise of the target compressed image to obtain a target image.

2. An image processing method according to claim 1, characterized in that: The resolution enlargement process of the image to be processed is to input the image to be processed into a pre-trained image super-resolution model for super-resolution reconstruction, and the resolution enlargement and compression process of the image to be processed are performed to obtain a target compressed image, including: Inputting the image to be processed into a pre-trained image super-resolution model for super-resolution reconstruction to obtain a target super-resolution image; The target super-resolution image is compressed to obtain the target compressed image in a preset format.

3. An image processing method according to claim 2, characterized in that: The image super-resolution model includes a pixel inverse rearrangement module, a first convolutional layer, a cyclic feature extraction module, a feature reconstruction module, a second convolutional layer, an upsampling layer and an output layer connected in sequence, wherein the cyclic feature extraction module is used to perform cyclic feature extraction processing on the first feature map output by the first convolutional layer to obtain a second feature map with enhanced feature representation capability, and the feature reconstruction module is used to perform feature reconstruction processing on the second feature map to obtain a feature reconstructed image with enlarged image detail information.

4. An image processing method according to claim 2, characterized in that: The image super-resolution model is pre-trained by the following steps: Acquire a plurality of preset super-resolution sample images, and perform image degradation processing on the super-resolution sample images to obtain low-resolution sample images; Generate a training data pair according to the low-resolution sample image and the corresponding super-resolution sample image; Inputting the training data pair into the initialized image super-resolution model to obtain a predicted super-resolution image corresponding to the low-resolution sample image; The model parameters of the image super-resolution model are updated according to the predicted super-resolution image and the super-resolution sample image to obtain the trained image super-resolution model.

5. An image processing method according to claim 4, characterized in that: The step of inputting the training data pair into the initialized image super-resolution model to obtain a predicted super-resolution image corresponding to the low-resolution sample image comprises: Performing image decomposition and pixel reorganization on the low-resolution sample image to obtain an inverse rearranged image; Performing convolution processing on the inverse rearranged image to obtain a first feature map; Performing a cyclic feature extraction process on the first feature map to obtain a second feature map; Performing feature reconstruction processing on the second feature map to obtain a feature reconstructed image; Performing convolution processing on the feature reconstructed image to obtain a third feature map, and then performing feature fusion processing on the first feature map and the third feature map to obtain a fourth feature map; The fourth feature map is up-sampled to obtain the predicted super-resolution image, and the predicted super-resolution image is output.

6. An image processing method according to claim 3, characterized in that: The performing cyclic feature extraction processing on the first feature map output by the first convolutional layer includes: Performing feature extraction processing on the first feature map to obtain a first extracted feature; Performing feature extraction on the obtained first extracted features and the first feature map together to obtain second extracted features; Performing feature extraction on the obtained first extracted features, the second extracted features, and the first feature map together to obtain a third extracted feature; Performing feature extraction on the obtained first extracted feature, the second extracted feature, ..., the i-th extracted feature and the first feature map together to obtain the i+1-th extracted feature; If the preset loop termination condition is not met, let i=i+1, where i is an integer greater than or equal to 3; When the preset loop termination condition is reached, the obtained (i+1)th extracted feature is used as the second feature map.

7. An image processing method according to claim 3, characterized in that: The cyclic feature extraction module includes a plurality of feature extraction layers, each of which includes a first convolution sublayer and a corresponding first activation function, and the cyclic feature extraction process is performed on the first feature map output by the first convolution layer, including: Performing convolution processing and activation operation processing on the first feature map through the first feature extraction layer to obtain a first activation map; Performing stacking processing on the obtained first activation map and the first feature map according to the convolution channel to obtain a first stacked feature map, and performing convolution processing and activation operation processing on the obtained first stacked feature map through the second feature extraction layer to obtain a second activation map; Performing stacking processing on the first activation map, the second activation map, and the first feature map obtained according to the convolution channel to obtain a second stacked feature map, and performing convolution processing and activation operation processing on the second stacked feature map obtained through the third feature extraction layer to obtain a third activation map; According to the convolution channel, the first activation map, the second activation map, ..., the j-th activation map and the first feature map are stacked to obtain a j-th stacked feature map, and the j-th stacked feature map is convolved and activated by the j+1-th feature extraction layer to obtain a j+1-th activation map; In the case where the j+1th feature extraction layer is not the last feature extraction layer, let j=j+1, where j is an integer greater than or equal to 3; When the j+1th feature extraction layer is the last feature extraction layer, the obtained j+1th activation map is used as the second feature map; or, when the j+1th feature extraction layer is the last feature extraction layer, the obtained first activation map, the second activation map, ..., the j+1th activation map and the first feature map are stacked according to the convolution channel to obtain the second feature map.

8. An image processing method according to claim 3, characterized in that: The feature reconstruction module includes a first normalization layer, a second convolution sublayer, a third convolution sublayer, a fourth convolution sublayer, a first activation sublayer, a fifth convolution sublayer, a second normalization layer, a sixth convolution sublayer, a seventh convolution sublayer, a second activation sublayer and an eighth convolution sublayer, and the feature reconstruction processing of the second feature map includes: Performing layer normalization processing on the second feature map through the first normalization layer to obtain a first feature submap; Through the second convolution sublayer, the third convolution sublayer and the fourth convolution sublayer, the first feature subgraph is convolved to obtain a second feature subgraph, a third feature subgraph and a fourth feature subgraph; Multiplying the third feature sub-graph and the fourth feature sub-graph to obtain a fifth feature sub-graph, and performing activation operation processing on the fifth feature sub-graph through the first activation sub-layer to obtain a second activation sub-graph; Multiplying the second feature subgraph and the second activation subgraph to obtain a sixth feature subgraph, and convolving the sixth feature subgraph through the fifth convolutional sublayer to obtain a seventh feature subgraph; performing an addition process on the second feature map and the seventh feature submap to obtain an eighth feature submap, and performing a layer normalization process on the eighth feature submap through the second normalization layer to obtain a ninth feature submap; Through the sixth convolution sublayer and the seventh convolution sublayer, convolution processing is performed on the ninth feature subgraph respectively to obtain a tenth feature subgraph and an eleventh feature subgraph; Performing activation operation processing on the tenth feature subgraph through the second activation sublayer to obtain a twelfth feature subgraph; The eleventh feature sub-image and the twelfth feature sub-image are multiplied to obtain a thirteenth feature sub-image, and the thirteenth feature sub-image is convolved through the eighth convolution sub-layer to obtain the feature reconstructed image.

9. The image processing method according to claim 3, characterized in that: The upsampling layer includes multiple ninth convolution sublayers and corresponding second activation functions and a tenth convolution sublayer, and the upsampling layer performs upsampling processing on the input fourth feature map through the following steps: Performing transposed convolution processing and activation operation processing on the fourth feature map input to the current layer through the ninth convolution sublayer of the current layer and the second activation function to obtain the upsampled fourth feature map, and then inputting the upsampled fourth feature map to the next layer; The tenth convolution sublayer performs transposed convolution processing on the input fourth feature map to obtain a predicted super-resolution image.

10. An image processing method according to claim 4, characterized in that: The updating of the model parameters of the image super-resolution model according to the predicted super-resolution image and the super-resolution sample image comprises: Calculating a loss value according to the predicted super-resolution image and the super-resolution sample image; According to the loss value, the model parameters of the image super-resolution model are updated through a back-propagation algorithm.

11. An image processing method according to claim 2, characterized in that: The compressing the target super-resolution image comprises: Performing color mode conversion on the target super-resolution image to obtain a first image; Performing chroma downsampling processing on the first image to obtain a second image; Performing block processing on the second image to obtain a plurality of first image matrices, and then performing discrete cosine transform on the first image matrix to obtain a second image matrix; Data quantization and Huffman coding are performed on the second image matrix to obtain the target compressed image.

12. An image processing method, characterized in that: Applied to the client, including: Uploading the image to be processed to the server, so that the server performs resolution enlargement and compression processing on the image to be processed to obtain a target compressed image; The target compressed image sent by the server is received, and compression noise of the target compressed image is removed to obtain a target image.

13. An image processing method according to claim 12, characterized in that: The step of removing compression noise of the target compressed image comprises: Determine a target grayscale image according to the target compressed image; Binarizing the target grayscale image to obtain a target binary image; Performing a morphological closing operation on the target binary image to obtain a target noise mask image; A noise region is determined according to the target noise mask image, and a mask is assigned to the noise region to obtain the target image.

14. An image processing system, characterized in that: It includes client and server, where: The client is configured to upload the image to be processed to the server, and receive the target compressed image sent by the server, remove the compression noise of the target compressed image, and obtain the target image; The server is configured to perform resolution enlargement and compression processing on the image to be processed to obtain a target compressed image, and send the target compressed image to the client.

15. A storage medium, the storage medium being a computer-readable storage medium, used for computer-readable storage, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the image processing method according to any one of claims 1 to 13.

Images