Image processing method, device, equipment and storage medium

Abstract

The present application provides an image processing method, apparatus, device and storage medium, the method comprising: obtaining an image to be processed and a convolution processing model, the convolution processing model being a trained model obtained from a server, downsampling the image to be processed to obtain a downsampled image, processing the downsampled image according to the convolution processing model to obtain a first transformation matrix set representing the degree of exposure, generating a target image according to the image to be processed and the first transformation matrix set, the dynamic range of the target image being higher than the dynamic range of the image to be processed. Through the present application, the generation of a high dynamic range image from a low dynamic range image is realized at the terminal, saving the time for obtaining a high dynamic range image and ensuring the security of the data.

Description

Technical Field

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

[0002] With the development of Internet technology, ultra-high-definition video technology has achieved vigorous development, and image acquisition and viewing devices have added support for high-dynamic range (HDR) images. Compared with traditional standard-dynamic range (SDR) images, HDR images retain the details of the brightness between the brightest and darkest parts of the image, more accurately describe the brightness difference, and present light and shadow in a way that is closer to what the human eye can perceive, thus providing viewers with a higher-quality viewing experience.

[0003] There are three main methods for generating HDR images: the first is to take multiple images with different exposures and merge them into one HDR image, which has the defect of being unable to process existing SDR images; the second is to transform the existing SDR images to obtain multiple images with different exposures, and then merge them into one HDR image, which is only suitable for images in specific scenes (such as night lights, building shadows, etc.); the third is to process the images through the neural network obtained by deep learning training to obtain HDR images. This method requires uploading the images to the server for processing, which is not only time-consuming, but also has data security issues. Summary of the invention

[0004] In view of this, embodiments of the present application provide an image processing method, apparatus, device, and storage medium.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present application provides an image processing method, including:

[0007] Obtaining an image to be processed and a convolution processing model, wherein the convolution processing model is a trained model obtained from a server;

[0008] Performing down-sampling processing on the image to be processed to obtain a down-sampled image;

[0009] Processing the downsampled image according to the convolution processing model to obtain a first transformation matrix set representing the exposure degree;

[0010] A target image is generated according to the image to be processed and the first transformation matrix set, wherein the dynamic range of the target image is higher than the dynamic range of the image to be processed.

[0011] The present application provides an image processing method, including:

[0012] Acquire training data, wherein the training data at least includes a reference image, a first set of training images having different exposure factors from the reference image, and a target reference image, wherein the target reference image is a high dynamic range image corresponding to the reference image;

[0013] Down-sampling the reference image and each first training image in the first training image set to obtain a corresponding reference down-sampled image and each first down-sampled image;

[0014] Training a preset convolution processing model based on the reference image, each first training image and the target reference image to obtain a trained convolution processing model;

[0015] Send the trained convolution processing model to the terminal.

[0016] The present application provides an image processing device, including:

[0017] A first acquisition module is used to acquire an image to be processed and a convolution processing model, where the convolution processing model is a trained model acquired from a server;

[0018] A first down-sampling module, used for performing down-sampling processing on the image to be processed to obtain a down-sampled image;

[0019] A processing module, configured to process the downsampled image according to the convolution processing model to obtain a first transformation matrix set representing the exposure degree;

[0020] A generating module is used to generate a target image according to the image to be processed and the first transformation matrix set, wherein the dynamic range of the target image is higher than the dynamic range of the image to be processed.

[0021] The present application provides an image processing device, including:

[0022] A second acquisition module is used to acquire training data, wherein the training data at least includes a reference image, a first set of training images having different exposure factors from the reference image, and a target reference image, wherein the target reference image is a high dynamic range image corresponding to the reference image;

[0023] A second down-sampling module is used to perform down-sampling processing on the reference image and each first training image in the first training image set to obtain a corresponding reference down-sampled image and each first down-sampled image;

[0024] A training module, used to train a preset convolution processing model based on the reference image, each first training image and a target reference image to obtain a trained convolution processing model;

[0025] A sending module is used to send the trained convolution processing model to the terminal.

[0026] The present application provides an image processing device, including:

[0027] A memory for storing executable instructions;

[0028] The processor is used to implement the method provided in the embodiment of the present application when executing the executable instructions stored in the memory.

[0029] An embodiment of the present application provides a storage medium storing executable instructions for causing a processor to execute and implement the method provided in the embodiment of the present application.

[0030] The present application provides an image processing method, apparatus, device and storage medium, wherein the method comprises: obtaining an image to be processed, downsampling the image to be processed to obtain a downsampled image, obtaining a trained convolution processing model from a server, processing the downsampled image according to the convolution processing model to obtain a first transformation matrix set characterizing the degree of exposure, and then generating a target image with a high dynamic range according to the image to be processed and the first transformation matrix set, by downsampling the high-resolution image to be processed and then performing transformation, generation and other processing on a terminal, a high dynamic range image can be generated on the terminal according to the low dynamic range image without uploading the low dynamic range image to be processed to the server, thereby saving time for obtaining the high dynamic range image and ensuring data security. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of the network architecture of the image processing method provided in an embodiment of the present application;

[0032] Figure 2 A schematic diagram of the structure of an image processing device 100 provided in an embodiment of the present application;

[0033] Figure 3 A schematic diagram of an implementation flow of the image processing method provided in an embodiment of the present application;

[0034] Figure 4 A schematic diagram of another implementation flow of the image processing method provided in an embodiment of the present application;

[0035] Figure 5 A schematic diagram of another implementation flow of the image processing method provided in the embodiment of the present application;

[0036] Figure 6 A schematic diagram of another implementation flow of the image processing method provided in the embodiment of the present application;

[0037] Figure 7 A schematic diagram of another implementation flow of the image processing method provided in the embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0039] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0040] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0041] 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 this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0042] In order to better understand the embodiments of the present application, the method of obtaining high dynamic range images in the related art and the existing shortcomings are first described.

[0043] With the development of Internet technology, ultra-high-definition video technology has flourished, and image acquisition and viewing devices have added support for high dynamic range (HDR). The "dynamic range" of an image refers to the ratio between the brightest and darkest parts of an image. The difference between the darkest and brightest objects that the human eye can perceive (dynamic range) is usually 10 12 , while traditional devices can only support up to 10 3 HDR technology expands the dynamic range to 10 5, allowing light and shadow to be presented in a way that is closer to what the human eye can perceive. Before HDR technology, SDR (Standard Dynamic Range, also referred to as LDR in some documents, SDR is used below) technology was used. SDR images map high-range brightness to low-range brightness, losing a lot of image details, making the brightest and darkest parts of the image blurry. Therefore, the technology of generating HDR images from SDR images came into being. There are currently several ways to obtain HDR images:

[0044] The first is to take several images with different exposures when the image is taken. For example, when shooting, take the following three images: an underexposed image, a correctly exposed image, and an overexposed image. Among them, the correctly exposed image is the regular image that the camera will take according to its estimated lighting; the underexposed image is darker than the correctly exposed image, and the goal is to capture the very bright part of the image; the overexposed image is brighter than the correctly exposed image, and the goal is to capture the very dark part of the image. The three images taken are combined into an HDR image through calculation steps such as alignment and merging.

[0045] HDR technology based on multiple image synthesis requires that the HDR mode be preset when the image is captured, multiple exposures are performed, multiple images of the same scene at different exposures are collected, and then these images are synthesized to generate an image with more brightness details. This method is limited to the generation of HDR images only when multiple exposures are preset when the image is collected, and it is impossible to process existing SDR images to generate HDR images.

[0046] The second method is to transform the brightness of an existing SDR image by calculation, scale the brightness range, or separate the illumination and reflected light of the image by filtering, and transform the brightness separately, so as to achieve the effect of supplementing the brightness details and removing the artifacts that should not be generated in the process. Alternatively, the existing SDR image is calculated to generate multiple images with different exposures, and then the first method is used to synthesize the HDR image.

[0047] The SDR image is processed based on image calculation. Since the brightness information of the SDR image is compressed, it is lost image information. Simply calculating and stretching the pixel brightness of the image cannot truly restore the lost brightness information. This method of generating HDR images can only generate good HDR images for images taken in specific scenes or lighting conditions (such as night lights, building shadows, etc.), and cannot be adapted to SDR images in any environment.

[0048] The third method is to use a trained deep learning neural network to process existing images and add brightness details. Based on deep learning, HDR generation is a method that uses a large amount of data training to train a suitable processing model for a large number of input SDR images and HDR images. Then, the SDR images are input into the trained processing model to calculate the HDR images. Based on a large amount of data training, this method can make the output HDR image less different from the target image of the received SDR image, and obtain a better HDR image.

[0049] However, the deep learning-based neural network processing method for generating HDR images requires computing on a server with strong computing power and cannot be applied to ordinary mobile terminals. When users generate HDR images, they need to upload SDR images to the server for processing, which is complicated and time-consuming. In addition, uploading SDR images to the server affects the user's data security.

[0050] Based on the above problems, an image processing method is provided in an embodiment of the present application, wherein a terminal obtains an image to be processed and downsamples it to obtain a downsampled image; a trained convolution processing model is obtained from a server, and the downsampled image is processed according to the convolution processing model to obtain a first transformation matrix set characterizing the degree of exposure, and then a target image with a high dynamic range is generated according to the image to be processed and the first transformation matrix set. By downsampling the high-resolution image to be processed and then performing transformation, generation and other processing on the terminal, a high dynamic range image can be generated on the terminal based on the low dynamic range image without uploading the low dynamic range image to be processed to the server, thereby saving time in obtaining the high dynamic range image and ensuring data security.

[0051] The following describes an exemplary application of the apparatus for implementing the embodiment of the present application. The apparatus provided in the embodiment of the present application can be implemented as a terminal device. The following describes an exemplary application of the terminal device when the apparatus is implemented as a terminal device.

[0052] See also Figure 1 , Figure 1 A schematic diagram of the network architecture of the image processing method provided in the embodiment of the present application is shown in FIG. Figure 1 As shown, the network architecture at least includes an image processing device 100, a server 200 and a network 300. To support an exemplary application, the image processing device 100 is connected to the server 200 via the network 300, which may be a local area network or a wide area network.

[0053] When a user needs to generate a high dynamic range HDR image based on a low dynamic range SDR image through the image processing device 100, the low dynamic range image to be processed is first obtained, and a request message is sent to the server 200 through the network 300, and the request message is used to request the acquisition of the convolution processing model. The server 200 determines and sends a response message to the image processing device 100 based on the obtained request information, and the response message includes the convolution processing model trained on the server. The image processing device 100 generates a high dynamic range image corresponding to the image to be processed based on the received convolution processing model and the obtained image to be processed, so that the user can obtain an image with more brightness and detail information.

[0054] The device provided in the embodiments of the present application may be implemented in the form of hardware or a combination of hardware and software. Various exemplary implementations of the device provided in the embodiments of the present application are described below.

[0055] according to Figure 2 The exemplary structure of the image processing device 100 shown in the figure can foresee other exemplary structures of the image processing device 100, so the structure described here should not be regarded as limiting. For example, some components described below can be omitted, or components not described below can be added to meet the special needs of certain applications.

[0056] Figure 2 The image processing device 100 shown includes: at least one processor 110, a memory 140, at least one network interface 120 and a user interface 130. Each component in the image processing device 100 is coupled together via a bus system 150. It is understood that the bus system 150 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 150 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 150 is not described in detail. Figure 2 Various buses are labeled as bus system 150 .

[0057] The user interface 130 may include a display, a keyboard, a mouse, a touch pad, a touch screen, and the like.

[0058] The memory 140 may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM). The volatile memory may be a random access memory (RAM). The memory 140 described in the embodiments of the present application is intended to include any suitable type of memory.

[0059] The memory 140 in the embodiment of the present application can store data to support the operation of the image processing device 100. Examples of such data include: any computer program for operating on the image processing device 100, such as an operating system and an application program. Among them, the operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application program can include various application programs.

[0060] As an example of the method provided in the embodiment of the present application being implemented by software, the method provided in the embodiment of the present application can be directly embodied as a combination of software modules executed by the processor 110. The software module can be located in a storage medium, and the storage medium is located in the memory 140. The processor 110 reads the executable instructions included in the software module in the memory 140, and completes the method provided in the embodiment of the present application in combination with the necessary hardware (for example, including the processor 110 and other components connected to the bus 150).

[0061] As an example, the processor 110 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0062] The image processing method provided in the embodiment of the present application will be described in conjunction with the exemplary application and implementation of the terminal provided in the embodiment of the present application.

[0063] See also Figure 3 , Figure 3 A schematic diagram of an implementation flow of the image processing method provided in the embodiment of the present application, which is applied to Figure 1 The image processing device in the network architecture shown in the figure may be a terminal, which includes at least a display screen, and can output a target image on the display screen. Figure 3 The steps shown are explained.

[0064] Step S301, obtaining an image to be processed and a convolution processing model.

[0065] The convolution processing model is a trained model obtained from a server.

[0066] When a user needs to generate an HDR target image, first obtain the SDR image corresponding to the pre-generated HDR target image, and determine the SDR image as the image to be processed. In practical applications, the image to be processed can be an image captured in real time by the user using the terminal's camera, or an image previously captured and saved in the terminal, or an image received by the terminal from other terminals. This embodiment does not specifically limit the source of the image to be processed. After the terminal obtains the image to be processed, it does not need to upload it to the server, ensuring the security of the data.

[0067] The terminal obtains the convolution processing model and first searches its own memory to see if the previously acquired convolution processing model is already stored. If not, the terminal obtains the convolution processing model from the server. The convolution processing model is a model obtained by the server through deep learning training using images of the same image with different exposure levels. The training operation of the convolution processing model is completed by the server, and the terminal can use it directly after receiving it from the server, making it possible to generate HDR images on the terminal side with low processing power.

[0068] The terminal can first obtain the SDR image, and then obtain the convolution processing model from the server when it needs to generate the HDR image. It can also obtain the convolution processing model first and then obtain the SDR image. This embodiment does not limit the order of obtaining. After obtaining the convolution processing model, the terminal saves the convolution processing model at the same time. When the user generates the HDR image next time, there is no need to repeatedly receive the convolution processing model from the server, which reduces the transmission data and shortens the time to generate the HDR image again.

[0069] Step S302: down-sample the image to be processed to obtain a down-sampled image.

[0070] Since the processing capability and processing speed of the terminal are not as good as those of the server, in order to generate the HDR target image on the terminal side, after obtaining the image to be processed, the terminal first downsamples the high-resolution image to be processed to obtain a low-resolution downsampled image.

[0071] Specifically, the resolution of the image to be processed is first determined, and the downsampling coefficient is determined according to the resolution and the preset resolution, wherein the preset resolution is the resolution of the pre-obtained downsampled image, and then the image to be processed is downsampled according to the downsampling coefficient to obtain a downsampled image.

[0072] In the specific implementation process, the size of the preset resolution can be predefined by the user. Generally, the preset resolution can be defined as 256×256. After downsampling, a low-resolution image with a resolution of 256×256 is obtained. The use of a low-resolution image can reduce the amount of calculation performed on the terminal, thereby speeding up the generation of HDR images on the terminal.

[0073] Step S303: Process the downsampled image according to the convolution processing model to obtain a first transformation matrix set representing the exposure degree.

[0074] Since the convolution processing model is a model obtained by the server through deep learning training based on images of the same image with different exposure levels, it has feature information of different exposure levels. The downsampled image is transformed according to the convolution processing model to obtain a series of first transformation matrices for exposure conversion of the downsampled image.

[0075] Each first transformation matrix in the first transformation matrix set represents the exposure level of an image, and the elements in the first transformation matrix represent the transformation coefficients of pixels at corresponding positions on the downsampled image.

[0076] Step S304: generating a target image according to the image to be processed and the first transformation matrix set.

[0077] The dynamic range of the target image is higher than the dynamic range of the image to be processed.

[0078] Here, the first transformation matrices in the first transformation matrix set obtained in the previous step are further processed to obtain multiple second transformation matrices whose sizes are equal to the resolution of the image to be processed. The corresponding pixels of the image to be processed are transformed according to each element of the multiple second transformation matrices to generate an HDR image corresponding to the image to be processed, which is determined to be the target image.

[0079] In the image processing method provided in the embodiment of the present application, the terminal obtains an image to be processed, downsamples the image to be processed to obtain a downsampled image, obtains a trained convolution processing model from a server, processes the downsampled image according to the convolution processing model, and obtains a first transformation matrix set characterizing the degree of exposure, and then generates a target image with a high dynamic range according to the image to be processed and the first transformation matrix set. By downsampling the high-resolution image to be processed and then performing transformation, generation, and other processing on the terminal, a high dynamic range image can be generated on the terminal based on the low dynamic range image without uploading the low dynamic range image to be processed to the server, thereby saving time in obtaining the high dynamic range image and ensuring data security.

[0080] In some embodiments, Figure 3 The step S304 "generating a target image according to the image to be processed and the first transformation matrix set" can be implemented by the following steps S3041 to S3043, which are described below in combination with each step.

[0081] Step S3041, performing grayscale processing on the image to be processed to generate a grayscale image containing edge brightness information.

[0082] Since the first transformation matrix is ​​obtained based on the low-resolution downsampled image, the size of the first transformation matrix is ​​the same as the resolution of the downsampled image, which is smaller than the resolution of the image to be processed. In order to obtain a target image with the same resolution as the image to be processed, it is necessary to perform upsampling processing on the first transformation matrix. The upsampling processing is guided by the image to be processed, so that the size of the second transformation matrix obtained after the upsampling processing is equal to the resolution of the target image.

[0083] Since HDR images are used to obtain more brightness details, they do not involve color feature information, that is, when upsampling, color feature information is unnecessary feature information. Based on this, before transforming the image to be processed, grayscale processing is first performed on it to generate a grayscale image containing edge brightness information, thereby removing unnecessary feature information in the image to be processed. The upsampling process is guided by the grayscale image after removing unnecessary feature information, thereby improving the processing speed of the terminal.

[0084] Step S3042: Generate a second transformation matrix set according to the grayscale image and the first transformation matrix set.

[0085] The size of each second transformation matrix in the second transformation matrix set is the same as the size of the image to be processed.

[0086] Under the guidance of the grayscale image that retains edge brightness information, the first transformation matrices with different exposure levels in the first transformation matrix set are upsampled respectively to obtain multiple second transformation matrices with the same resolution as the grayscale image, and the second transformation matrix set is determined by the multiple second transformation matrices.

[0087] A specific implementation method for generating a second transformation matrix set based on a grayscale image and a first transformation matrix set is: first, based on each first transformation matrix in the first transformation matrix set and the grayscale image, determine the interpolation information corresponding to each first transformation matrix, and then interpolate the each first transformation matrix according to the each interpolation information to generate each second transformation matrix, thereby obtaining a second transformation matrix set.

[0088] Step S3043: transform the image to be processed according to the second transformation matrix set to generate a target image.

[0089] The elements of the second transformation matrix in the second transformation matrix set represent the transformation coefficients of the pixels at the corresponding positions on the image to be processed. The second transformation matrix set is used to transform the image to be processed to obtain intermediate images with different exposures, and then these intermediate images with different exposures are merged to obtain the target image.

[0090] This embodiment guides the first transformation matrix set to generate a second transformation matrix set having the same matrix size as the image to be processed based on the grayscale image obtained after grayscale processing of the image to be processed, thereby accelerating the processing speed of the terminal to generate the target image.

[0091] In some embodiments, the above-mentioned step S3043 "transform the image to be processed according to the second transformation matrix set to generate a target image" can be implemented through the following steps S30431 to S30432, which are explained below in combination with each step.

[0092] Step S30431, transform each pixel of the image to be processed one by one according to each second transformation matrix in the second transformation matrix set, and obtain corresponding intermediate images.

[0093] The elements of the second transformation matrices in the second transformation matrix set represent the transformation coefficients of the pixels at the corresponding positions on the image to be processed. According to the transformation coefficients of the corresponding positions of each second transformation matrix, each pixel of the image to be processed is transformed one by one to obtain intermediate images of different exposure levels corresponding to each second transformation matrix.

[0094] Step S30432, performing fusion processing on the intermediate images to generate a target image.

[0095] Among the multiple intermediate images with different exposure levels, the weak light part of the image with a high exposure level and the strong light part of the image with a low exposure level are selected and synthesized to obtain the target image.

[0096] Specifically, the exposure coefficients of the intermediate images are first obtained, and then based on the exposure coefficients of the intermediate images, a first target intermediate image and a second target intermediate image are determined from the intermediate images, wherein the exposure coefficient of the first target intermediate image is greater than a first exposure threshold, and the exposure coefficient of the second target intermediate image is less than a second exposure threshold. The first exposure threshold is greater than or equal to the second exposure threshold, and the first target intermediate image is determined to be an image with a high exposure degree, which improves the overall brightness of the image to be processed, and the second target intermediate image is an image with a low exposure degree, which reduces the overall brightness of the image to be processed.

[0097] After obtaining the first target intermediate image and the second target intermediate image with different exposure levels, a first image area is determined from the first target intermediate image, and a second image area is determined from the second target intermediate image, wherein the brightness of the first image area is less than a first brightness threshold, and the brightness of the second image area is greater than a second brightness threshold. The first brightness threshold is less than or equal to the second brightness threshold, and the determined first image area corresponds to a darker area in the image to be processed, and the second image area corresponds to a brighter area in the image to be processed.

[0098] Finally, the first image area and the second image area are merged to obtain a target image. The target image is obtained by merging the weak light area in the first target intermediate image with increased brightness and the strong light area in the second target intermediate image with decreased brightness.

[0099] When there are multiple first intermediate target images and second intermediate target images, the merging weight is first determined according to the exposure coefficient, and then they are merged into one target image. If the merged area of ​​the first image area and the second image area is smaller than the image to be processed, it indicates that there is a third intermediate target image whose brightness is within the first brightness threshold and the second brightness threshold. When generating the target image, it is not necessary to process the image area of ​​the third intermediate target image, but when merging, it is necessary to merge the area to obtain a completed target image.

[0100] This embodiment transforms the image to be processed according to the second transformation matrix set to obtain multiple intermediate images with different exposure levels, selects the weak light area in the intermediate image with a higher exposure level and the strong light area in the intermediate image with a lower exposure level to synthesize them to obtain a target image, thereby realizing the generation of a high dynamic range image based on a low dynamic range image on the terminal.

[0101] In some embodiments, Figure 3 The step S303 shown, "processing the downsampled image according to the convolution processing model to obtain a matrix set representing the exposure degree" can be implemented through the following steps S3031 to S3032, which are explained below in combination with each step.

[0102] Step S3031, extracting local feature information and global feature information of the down-sampled image.

[0103] Step S3032: input the local feature information and the global feature information into the convolution processing model to obtain a first transformation matrix set representing the exposure degree.

[0104] In this embodiment, the existing feature extraction method can be used to extract the local feature information and global feature information of the downsampled image, and the extracted information is input as input data to the convolution processing model to determine the first transformation matrix set. Since the convolution processing model is trained by training data of multiple different exposure levels, multiple first transformation matrices of different exposure levels can be obtained after inputting the extracted information. Then, the first transformation matrix set representing the exposure level is obtained from these first transformation matrices, so that multiple images of different exposure levels of the downsampled image corresponding to the image to be processed can be obtained. The calculation is simple and can be performed on a terminal with low processing power, making it possible to generate HDR images on the terminal.

[0105] See also Figure 4 , Figure 4 Another implementation flow diagram of the image processing method provided in the embodiment of the present application is applied to Figure 1 The server in the network architecture shown will be combined with Figure 4 The steps shown are explained.

[0106] Step S401, obtaining training data.

[0107] The training data at least includes a reference image, a first set of training images having different exposure factors from the reference image, and a target reference image, wherein the target reference image is a high dynamic range image corresponding to the reference image.

[0108] Step S402: down-sample the reference image and each first training image in the first training image set to obtain a corresponding reference down-sampled image and each first down-sampled image.

[0109] The high-resolution training images are downsampled to obtain low-resolution training images, which are then used as input data for training, thereby reducing the amount of training data and improving the training speed.

[0110] Step S403: training a preset convolution processing model based on the reference image, each first training image and the target reference image to obtain a trained convolution processing model.

[0111] The convolution processing model is trained by performing deep learning on the server using the same image at different exposure levels.

[0112] In this embodiment, the loss function used in training is a loss calculation performed on a high-resolution training image before downsampling. When the loss reaches an expected range, it is determined that the convolution processing model training is completed. At this time, the difference between the predicted image obtained according to the trained convolution processing model and the actual HDR image is small, which meets the user's requirements.

[0113] Step S404: sending the trained convolution processing model to the terminal.

[0114] Here, the server may send the convolution processing model after receiving a request message for obtaining the convolution processing model, or may send the trained convolution processing model to the terminal after successfully connecting to the terminal.

[0115] In the image processing method provided in the embodiment of the present application, the server obtains training data, performs downsampling processing on the reference image and each first training image in the training data, obtains the corresponding reference downsampling image and each first downsampling image, and then trains the preset convolution processing model based on the reference image, each first training image and the target reference image to obtain the trained convolution processing model, and sends the trained convolution processing model to the terminal to generate an HDR target image on the terminal. By training the convolution processing model on a server with strong processing power, and sending the trained convolution processing model to the terminal after the training is completed, it is possible to generate an HDR target image on a terminal with low processing power, and the server does not need to receive the low dynamic range image sent by the terminal, thereby ensuring the security of the data.

[0116] In some embodiments, Figure 4 The step S403 shown in FIG. 4 "training a preset convolution processing model based on the reference image, each first training image and the target reference image to obtain a trained convolution processing model" can be implemented through the following steps S4031 to S4033, which are explained below in combination with each step.

[0117] Step S4031, extracting local feature information and global feature information of the reference image and each first training image respectively.

[0118] Here, existing feature extraction methods may be used to extract partial features and global features from training images with different exposure levels, so as to obtain local feature information and global feature information of the reference image and each first training image.

[0119] Step S4032: input the local feature information and the global feature information of the reference image and each first training image into a preset convolution processing model to obtain a predicted image.

[0120] The local feature information and global feature information extracted in the previous step are used as training input data and input into the preset convolution processing model for training. The output data obtained is the predicted image.

[0121] Step S4033, using the predicted image and the target reference image to adjust the parameters of the convolution processing model to obtain a trained convolution processing model.

[0122] The loss of the predicted image and the real target reference image corresponding to the reference image is estimated. When the loss is too large and exceeds the preset loss range, the convolution processing model is adjusted, and the local feature information and the global feature information are used as input data to obtain a new predicted image. The training is terminated when the loss is within the preset range, thereby obtaining a trained convolution processing model.

[0123] This embodiment reduces the amount of data calculated when the terminal generates an HDR image by performing training on a server with strong processing capabilities. Compared with training performed on the terminal, the training speed is increased, thereby speeding up the processing speed of the terminal generating HDR images.

[0124] Based on the above embodiments, the present application further provides an image processing method, which is applied to Figure 1 The network architecture shown, Figure 5 A schematic diagram of another implementation flow of the image processing method provided in the embodiment of the present application is shown as follows: Figure 5 As shown, the method includes:

[0125] Step S501: The server obtains training data.

[0126] The training data at least includes a reference image, a first set of training images having different exposure factors from the reference image, and a target reference image, wherein the target reference image is a high dynamic range image corresponding to the reference image.

[0127] Step S502: The server performs down-sampling processing on the reference image and each first training image in the first training image set to obtain a corresponding reference down-sampled image and each first down-sampled image.

[0128] The high-resolution training images are downsampled to obtain low-resolution training images, which are then used as input data for training, thereby reducing the amount of training data and improving the training speed.

[0129] Step S503: The server extracts local feature information and global feature information of the reference image and each first training image respectively.

[0130] Step S504: The server inputs the local feature information and the global feature information of the reference image and each first training image into a preset convolution processing model to obtain a predicted image.

[0131] The local feature information and global feature information of the downsampled image are used as training input data and input into a preset convolution processing model for training. The output data obtained is the predicted image.

[0132] Step S505: The server adjusts the parameters of the convolution processing model using the predicted image and the target reference image to obtain a trained convolution processing model.

[0133] The goal of deep learning training on the server side is to model the brightness transformation of the same image at different exposure levels, estimate the loss based on the pixel difference between the predicted image and the actual target reference image, and obtain a convolution processing model with pixel differences within an acceptable range through training.

[0134] Step S506: the terminal obtains the image to be processed.

[0135] The image to be processed is an SDR image corresponding to the pre-generated HDR target image. After the terminal obtains the image to be processed, it does not need to upload it to the server, ensuring the security of the data.

[0136] Step S507: the terminal performs down-sampling processing on the image to be processed to obtain a down-sampled image.

[0137] In order to generate an HDR target image on a terminal with low processing power, after obtaining the image to be processed, the terminal first downsamples the high-resolution image to be processed to obtain a low-resolution downsampled image to reduce the amount of data calculated by the terminal.

[0138] Step S508: The terminal obtains a convolution processing model.

[0139] Here, the convolution processing model is a trained model obtained from a server.

[0140] Step S509: The terminal extracts local feature information and global feature information of the down-sampled image.

[0141] In this embodiment, the existing feature extraction method can be used to extract the local feature information and the global feature information of the down-sampled image, which will not be described in detail here.

[0142] In step S510, the terminal inputs the local feature information and the global feature information into the convolution processing model to obtain a first transformation matrix set representing the exposure degree.

[0143] Since the convolution processing model is trained by training data of multiple exposure levels, after the extracted information is input as input data into the convolution processing model, multiple first transformation matrices of different exposure levels can be obtained.

[0144] Step S511: the terminal performs grayscale processing on the image to be processed to generate a grayscale image containing edge brightness information.

[0145] Before transforming the image to be processed, grayscale processing is first performed on it to generate a grayscale image containing edge brightness information, thereby removing unnecessary feature information from the image to be processed. The upsampling process is guided by the grayscale image after removing the unnecessary feature information, thereby improving the processing speed of the terminal.

[0146] Step S512: The terminal determines interpolation information corresponding to each first transformation matrix according to each first transformation matrix in the first transformation matrix set and the grayscale image.

[0147] In this embodiment, interpolation is used for upsampling. Specifically, the grayscale image is used as a guide image to determine a specific interpolation that meets the characteristics of the image to be processed, and each first transformation matrix is ​​interpolated and upsampled according to the determined interpolation.

[0148] Step S513: The terminal performs interpolation processing on each of the first transformation matrices according to each of the interpolation information to generate each of the second transformation matrices to obtain a second transformation matrix set.

[0149] The size of each second transformation matrix in the second transformation matrix set is the same as the size of the image to be processed.

[0150] In step S514, the terminal transforms each pixel of the image to be processed one by one according to each second transformation matrix in the second transformation matrix set, and obtains corresponding intermediate images.

[0151] The elements of the second transformation matrices in the second transformation matrix set represent the transformation coefficients of the pixels at the corresponding positions on the image to be processed. According to the transformation coefficients of the corresponding positions of each second transformation matrix, each pixel of the image to be processed is transformed one by one to obtain intermediate images of different exposure levels corresponding to each second transformation matrix.

[0152] Step S515: the terminal obtains the exposure coefficient of each intermediate image.

[0153] Step S516: The terminal determines a first target intermediate image and a second target intermediate image from the intermediate images based on the exposure coefficients of the intermediate images.

[0154] Among them, the exposure coefficient of the first target intermediate image is greater than the first exposure threshold, the exposure coefficient of the second target intermediate image is less than the second exposure threshold, and the first exposure threshold is greater than or equal to the second exposure threshold.

[0155] The determined first target intermediate image is an image with a high exposure level, which improves the overall brightness of the image to be processed, and the determined second target intermediate image is an image with a low exposure level, which reduces the overall brightness of the image to be processed.

[0156] In step S517, the terminal determines a first image area from the first target intermediate image, and determines a second image area from the second target intermediate image.

[0157] The brightness of the first image area is less than a first brightness threshold, the brightness of the second image area is greater than a second brightness threshold, and the first brightness threshold is less than or equal to the second brightness threshold.

[0158] The determined first image region corresponds to a region with a relatively dark brightness in the image to be processed, and the determined second image region corresponds to a region with a relatively bright brightness in the image to be processed.

[0159] Step S518: The terminal merges the first image area and the second image area to obtain a target image.

[0160] The dynamic range of the target image is higher than the dynamic range of the image to be processed.

[0161] A target image with a higher dynamic range than the image to be processed is obtained by merging the weak light area in the first target intermediate image with improved brightness and the strong light area in the second target intermediate image with reduced brightness.

[0162] The image processing method provided in the embodiment of the present application trains a convolution processing model according to training data on a server, sends the trained convolution processing model to a terminal, and after the terminal obtains the image to be processed, downsamples it to obtain a downsampled image, processes the downsampled image according to the obtained convolution processing model to obtain a first transformation matrix set characterizing the degree of exposure, and then generates a target image with a high dynamic range according to the image to be processed and the first transformation matrix set. By downsampling the high-resolution image to be processed and then performing transformation, generation and other processing on the terminal, a high dynamic range image can be generated on the terminal according to the low dynamic range image without uploading the low dynamic range image to be processed to the server, thereby saving time for obtaining the high dynamic range image and ensuring data security.

[0163] The following is an explanation of an exemplary application of the embodiments of the present application in a practical application scenario.

[0164] The present application provides an image processing method. Figure 6 A schematic diagram of another implementation flow of the image processing method provided in the embodiment of the present application is shown as follows: Figure 6 As shown, the method includes:

[0165] Step S601: The first acquisition module of the terminal acquires the image to be processed input by the user. For example, the image to be processed input by the user is a high-resolution image with a resolution of 1080p.

[0166] Step S602: The downsampling module downsamples the image to be processed to obtain a low-resolution downsampled image. The 1080p high-resolution image to be processed is downsampled to obtain a 480p low-resolution downsampled image.

[0167] Step S603: The first feature extraction module performs convolution processing on the low-resolution downsampled image to extract feature values.

[0168] Step S604: The first transformation matrix module performs calculations according to the extracted eigenvalues ​​and the convolution processing models under different exposure rates obtained from the server to obtain a first transformation matrix set.

[0169] By taking the global feature information and local feature information of a low-resolution downsampled image as input data and inputting it into a trained convolution processing model obtained from a server, a first transformation matrix set {A} of images with the same resolution but different exposure levels as the low-resolution image is obtained, and each first transformation matrix corresponds to an exposure coefficient.

[0170] Step S605: the grayscale processing module generates a grayscale image retaining edge information according to the image to be processed.

[0171] Step S606: The second transformation matrix module performs transformation processing according to the grayscale image and the first transformation matrix set to obtain a series of second transformation matrices equal to the resolution of the image to be processed, that is, the second transformation matrix set.

[0172] The edge brightness information of the image to be processed is recorded through a grayscale image, which guides the upsampling process of the first transformation matrix set {A} to obtain a second transformation matrix set {A'} having the same resolution size as the image to be processed.

[0173] Step S607: the processing module uses the second transformation matrix set to perform pixel-by-pixel processing on the image to be processed to obtain a series of multiple images with different exposures corresponding to the image to be processed.

[0174] By combining the image to be processed and the second transformation matrix set {A'} corresponding to different exposure levels, a plurality of images with different exposure levels corresponding to the image to be processed are calculated.

[0175] Step S608: The synthesis module synthesizes multiple images with different exposures to obtain a high dynamic range image.

[0176] Multiple images with different exposure levels are merged, and processes such as de-duplication, denoising, and brightness selection are performed to obtain a high dynamic range image with the same resolution as the image to be processed (i.e., 1080p).

[0177] The image processing method provided in this embodiment is an HDR picture generation method based on deep learning. First, a high-resolution image to be processed is downsampled so that most of the subsequent image operations are based on low-resolution images. The acquisition server is trained on images with different exposure levels to obtain a convolution processing model that can predict different exposure coefficients of a single image, so as to obtain first transformation matrices of images with different exposures in combination with the downsampled images. Then, each first transformation matrix is ​​upsampled guided by edge brightness information of the image to be processed to obtain second transformation matrices that can be applied to the image to be processed, so as to generate multiple images with different exposures. The generated multiple images with different exposures are then synthesized to achieve the purpose of obtaining details of different light brightness on the image and generate a high dynamic range image, so that brightness restoration processing of low dynamic range images is realized on terminals with limited computing and processing capabilities (such as smart phones, tablet computers, etc.). Compared with the traditional method of generating HDR images, the computational complexity is reduced.

[0178] The present application provides an image processing method. Figure 7 A schematic diagram of another implementation flow of the image processing method provided in the embodiment of the present application is shown as follows: Figure 7 As shown, the method includes:

[0179] Step S701: The second acquisition module of the server acquires training data.

[0180] The SDR images included in the training data include multiple images of the same image at different exposure levels.

[0181] Step S702: The second feature extraction module performs multiple convolution processes on the downsampled SDR image to extract global feature information and local feature information.

[0182] In step S703, the training module inputs the extracted global feature information and local feature information into the training model for training to obtain a predicted image, uses the real target image included in the training data to perform loss estimation, and terminates the training when the loss is less than a preset value to obtain a trained convolution processing model.

[0183] Step S704: The sending module sends the trained convolution processing model to the terminal.

[0184] The following is a description of an exemplary structure of the image processing device 80 provided in the embodiment of the present application implemented as a software module. In some embodiments, Figure 2 As shown, the software modules stored in the image processing device 80 in the memory 140 may include:

[0185] The first acquisition module 81 is used to acquire the image to be processed and the convolution processing model.

[0186] Here, the convolution processing model is a trained model obtained from a server.

[0187] The first down-sampling module 82 is used to perform down-sampling processing on the image to be processed to obtain a down-sampled image.

[0188] The processing module 83 is used to process the downsampled image according to the convolution processing model to obtain a first transformation matrix set representing the exposure degree.

[0189] The generating module 84 is used to generate a target image according to the image to be processed and the first transformation matrix set.

[0190] The dynamic range of the target image is higher than the dynamic range of the image to be processed.

[0191] In some embodiments, the processing module 83 further comprises:

[0192] The first extraction submodule is used to extract local feature information and global feature information of the down-sampled image.

[0193] The first input submodule is used to input the local feature information and the global feature information into the convolution processing model to obtain a first transformation matrix set representing the exposure degree.

[0194] In some embodiments, the generating module 84 further comprises:

[0195] The grayscale processing submodule is used to perform grayscale processing on the image to be processed to generate a grayscale image containing edge brightness information.

[0196] A generating submodule is used to generate a second transformation matrix set according to the grayscale image and the first transformation matrix set.

[0197] The size of each second transformation matrix in the second transformation matrix set is the same as the size of the image to be processed.

[0198] The transformation submodule is used to transform the image to be processed according to the second transformation matrix set to generate a target image.

[0199] In some embodiments, the generating submodule further comprises:

[0200] A determination unit is used to determine interpolation information corresponding to each first transformation matrix according to each first transformation matrix in the first transformation matrix set and the grayscale image.

[0201] The interpolation unit is used to perform interpolation processing on the first transformation matrices according to the interpolation information to generate second transformation matrices and obtain a second transformation matrix set.

[0202] In some embodiments, the transformation submodule further comprises:

[0203] The transformation unit is used to transform each pixel of the image to be processed one by one according to each second transformation matrix in the second transformation matrix set, so as to obtain corresponding intermediate images.

[0204] The fusion unit is used to fuse the intermediate images to generate a target image.

[0205] In some embodiments, the fusion unit further comprises:

[0206] The acquisition subunit is used to acquire the exposure coefficient of each intermediate image.

[0207] The first determination subunit is used to determine a first target intermediate image and a second target intermediate image from the intermediate images based on the exposure coefficients of the intermediate images.

[0208] The exposure coefficient of the first target intermediate image is greater than the first exposure threshold, and the exposure coefficient of the second target intermediate image is less than the second exposure threshold.

[0209] The second determining subunit is used to determine a first image area from the first target intermediate image and to determine a second image area from the second target intermediate image.

[0210] The brightness of the first image area is less than a first brightness threshold, and the brightness of the second image area is greater than a second brightness threshold.

[0211] The merging subunit is used to merge the first image area and the second image area to obtain a target image.

[0212] The present application further provides an image processing device, the image processing device comprising at least:

[0213] The second acquisition module is used to acquire training data.

[0214] The training data at least includes a reference image, a first set of training images having different exposure factors from the reference image, and a target reference image, wherein the target reference image is a high dynamic range image corresponding to the reference image.

[0215] The second down-sampling module is used to perform down-sampling processing on the reference image and each first training image in the first training image set to obtain a corresponding reference down-sampled image and each first down-sampled image.

[0216] The training module is used to train the preset convolution processing model based on the reference image, each first training image and the target reference image to obtain a trained convolution processing model.

[0217] A sending module is used to send the trained convolution processing model to the terminal.

[0218] In some embodiments, the training module further comprises:

[0219] The second extraction submodule is used to extract local feature information and global feature information of the reference image and each first training image respectively.

[0220] The second input submodule is used to input the local feature information and the global feature information of the reference image and each first training image into a preset convolution processing model to obtain a predicted image.

[0221] The adjustment submodule is used to adjust the parameters of the convolution processing model using the predicted image and the target reference image to obtain a trained convolution processing model.

[0222] The present application embodiment provides a storage medium storing executable instructions. When the executable instructions are executed by a processor, the processor will execute the method provided by the present application embodiment, for example, Figure 3 , Figure 4 and Figure 5 The method shown.

[0223] In some embodiments, the storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or various devices including one or any combination of the above memories.

[0224] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0225] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0226] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0227] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An image processing method, characterized in that: include: Obtaining an image to be processed and a convolution processing model, wherein the convolution processing model is a trained model obtained from a server; Performing down-sampling processing on the image to be processed to obtain a down-sampled image; Extracting local feature information and global feature information of the downsampled image; Inputting the local feature information and the global feature information into the convolution processing model to obtain a first transformation matrix set representing the exposure degree; Generate a target image according to the image to be processed and the first transformation matrix set, wherein the dynamic range of the target image is higher than the dynamic range of the image to be processed; The step of generating a target image according to the image to be processed and the first transformation matrix set includes: Performing grayscale processing on the image to be processed to generate a grayscale image containing edge brightness information; Generate a second transformation matrix set according to the grayscale image and the first transformation matrix set, wherein the size of each second transformation matrix in the second transformation matrix set is the same as the size of the image to be processed; Performing transformation processing on the image to be processed according to the second transformation matrix set to obtain various intermediate images; Obtaining the exposure factor of each intermediate image; Determine a first target intermediate image and a second target intermediate image from the intermediate images based on exposure coefficients of the intermediate images, wherein the exposure coefficient of the first target intermediate image is greater than a first exposure threshold, and the exposure coefficient of the second target intermediate image is less than a second exposure threshold; Determine a first image region from the first target intermediate image, and determine a second image region from the second target intermediate image, wherein brightness of the first image region is less than a first brightness threshold, and brightness of the second image region is greater than a second brightness threshold; The first image region and the second image region are merged to obtain a target image.

2. The method according to claim 1, characterized in that The step of generating a second transformation matrix set according to the grayscale image and the first transformation matrix set includes: Determine interpolation information corresponding to each first transformation matrix according to each first transformation matrix in the first transformation matrix set and the grayscale image; Interpolation processing is performed on each of the first transformation matrices according to each of the interpolation information to generate each of the second transformation matrices to obtain a second transformation matrix set.

3. The method according to claim 1, characterized in that The step of transforming the image to be processed according to the second transformation matrix set to obtain each of the intermediate images includes: Each pixel of the image to be processed is transformed one by one according to each second transformation matrix in the second transformation matrix set, so as to obtain corresponding intermediate images.

4. An image processing method, characterized in that: include: Acquire training data, wherein the training data at least includes a reference image, a first set of training images having different exposure factors from the reference image, and a target reference image, wherein the target reference image is a high dynamic range image corresponding to the reference image; Down-sampling the reference image and each first training image in the first training image set to obtain a corresponding reference down-sampled image and each first down-sampled image; Respectively extracting local feature information and global feature information of the corresponding reference down-sampled image and each first down-sampled image; Inputting the local feature information and the global feature information of the corresponding reference down-sampled image and each first down-sampled image into a preset convolution processing model to obtain a predicted image; Using the predicted image and the target reference image to adjust the parameters of the convolution processing model to obtain a trained convolution processing model; Send the trained convolution processing model to the terminal.

5. An image processing device, characterized in that: include: A first acquisition module is used to acquire an image to be processed and a convolution processing model, where the convolution processing model is a trained model acquired from a server; A first down-sampling module, used for performing down-sampling processing on the image to be processed to obtain a down-sampled image; A processing module, used for extracting local feature information and global feature information of the down-sampled image, inputting the local feature information and the global feature information into the convolution processing model, and obtaining a first transformation matrix set representing the exposure degree; A generating module, configured to generate a target image according to the image to be processed and the first transformation matrix set, wherein the dynamic range of the target image is higher than the dynamic range of the image to be processed; Wherein, the generation module is specifically used for: Performing grayscale processing on the image to be processed to generate a grayscale image containing edge brightness information; Generate a second transformation matrix set according to the grayscale image and the first transformation matrix set, wherein the size of each second transformation matrix in the second transformation matrix set is the same as the size of the image to be processed; Performing transformation processing on the image to be processed according to the second transformation matrix set to obtain various intermediate images; Obtaining the exposure factor of each intermediate image; Determine a first target intermediate image and a second target intermediate image from the intermediate images based on exposure coefficients of the intermediate images, wherein the exposure coefficient of the first target intermediate image is greater than a first exposure threshold, and the exposure coefficient of the second target intermediate image is less than a second exposure threshold; Determine a first image region from the first target intermediate image, and determine a second image region from the second target intermediate image, wherein brightness of the first image region is less than a first brightness threshold, and brightness of the second image region is greater than a second brightness threshold; The first image region and the second image region are merged to obtain a target image.

6. An image processing device, characterized in that: include: A second acquisition module is used to acquire training data, wherein the training data at least includes a reference image, a first set of training images having different exposure factors from the reference image, and a target reference image, wherein the target reference image is a high dynamic range image corresponding to the reference image; A second down-sampling module is used to perform down-sampling processing on the reference image and each first training image in the first training image set to obtain a corresponding reference down-sampled image and each first down-sampled image; A training module, used to extract local feature information and global feature information of the corresponding reference down-sampled image and each first down-sampled image, input the local feature information and global feature information of the corresponding reference down-sampled image and each first down-sampled image into a preset convolution processing model to obtain a predicted image, and adjust the parameters of the convolution processing model using the predicted image and the target reference image to obtain a trained convolution processing model; A sending module is used to send the trained convolution processing model to the terminal.

7. An image processing device, characterized in that: include: A memory for storing executable instructions; A processor, configured to implement the method according to any one of claims 1 to 4 when executing the executable instructions stored in the memory.

8. A storage medium, characterized in that: Executable instructions are stored, which are used to cause a processor to execute and implement the method described in any one of claims 1 to 4.

Images