Image quality optimization method and system, electronic equipment and storage medium

Abstract

The invention discloses an image quality optimization method and system, electronic equipment and a storage medium. The method comprises the following steps: acquiring a current target image acquired by a camera; performing multi-term image preprocessing on the current target image; inputting the preprocessed current target image into a pre-trained image processing model, identifying feature information of the current target image in each target dimension through the image processing model, and determining categories of the current target image in multiple dimensions based on the feature information, so as to obtain the target image based on the feature information and each category. Outputting a plurality of optimization parameters; the image processing model is obtained through training by taking a first sample image and a second sample image collected in the same scene as input in advance; the first sample image and the second sample image are respectively images acquired when rearview mirror glass is not arranged in front of the camera; and optimizing the current target image by using the multiple optimization parameters of the current target image to obtain an optimized current target image.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of image processing technology, and in particular to an image quality optimization method and system, an electronic device, and a storage medium. Background Art

[0002] With the development of automobile manufacturing and digital image processing technology, cameras are increasingly used in automobiles, and their application scenarios are becoming more and more diverse. Currently, in addition to cameras installed on the outside of the car for obstacle detection and intelligent assisted driving, more and more cameras are also installed inside the car's cabin to realize many functions and applications based on driver and passenger behavior detection.

[0003] In order to solve the problems of installation, space and aesthetics, the RGB-IR camera in the car cabin has begun to be installed inside the car rearview mirror, that is, it is installed in the housing of the rearview mirror in the car and hidden behind the rearview mirror glass in the car. This installation method makes full use of the internal space of the rearview mirror housing and avoids the camera being exposed to the outside. In addition, the rearview mirror glass can still be used normally, and the RGB-IR camera can also pass through the rearview mirror glass.

[0004] However, since the rearview mirror glass is for the driver to observe the situation behind the vehicle, it is different from ordinary glass and will be coated, which makes its light transmittance and reflectivity inferior to ordinary glass. Therefore, the image captured by the camera will have obvious color shadows and the white balance is difficult to calibrate. That is, the quality of the captured image is relatively low, which affects subsequent use. Summary of the invention

[0005] Based on the above-mentioned deficiencies of the prior art, the present application provides an image quality optimization method and system, an electronic device, and a storage medium to solve the problem of relatively low quality of existing captured images.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] The first aspect of the present application provides an image quality optimization method, comprising:

[0008] Get the current target image captured by the camera;

[0009] Performing multiple image preprocessing on the current target image;

[0010] The preprocessed current target image is input into a pre-trained image processing model, and the input current target image is processed by the image processing model to obtain multiple optimization parameters of the current target image; wherein the image processing model identifies feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on each item of feature information of the current target image, so as to output multiple optimization parameters of the current target image based on each item of feature information of the current target image and each category of the current target image; the image processing model is trained in advance using a first sample image and a second sample image collected in the same scene as input; the first sample image is an image collected when a rearview mirror glass is set in front of the camera; the second sample image is an image collected when no rearview mirror glass is set in front of the camera;

[0011] The current target image is optimized using a plurality of optimization parameters of the current target image to obtain the optimized current target image.

[0012] Optionally, in the above-mentioned image quality optimization method, before obtaining the current target image captured by the camera, it also includes:

[0013] After the system is powered on, the camera is configured to control the camera to start real-time image acquisition and output based on the configured information.

[0014] Optionally, in the above-mentioned image quality optimization method, the performing multiple image preprocessing steps on the current target image includes:

[0015] The current target image is processed using each of the image preprocessing methods in accordance with the processing order of the image preprocessing methods specified by each preset item; wherein each of the image preprocessing methods includes multiple items of automatic exposure control, black level correction, bad pixel correction, RAW noise reduction, lens shading correction, automatic white balance, color interpolation, color matrix correction, gamma curve adjustment, edge enhancement and color space conversion.

[0016] Optionally, in the above-mentioned image quality optimization method, the processing of the current target image by using each of the image preprocessing methods according to the preset processing order of the image preprocessing methods includes:

[0017] Calculating the difference between the brightness of the current target image and the preset target brightness, and using the difference to adjust the exposure parameters of the camera;

[0018] Using a preset fixed value, performing black level correction on the current target image;

[0019] Detecting bad pixels in the current target image through gradients in multiple directions, and correcting the bad pixels;

[0020] Filtering out noise information in the current target image by using a filter;

[0021] Performing lens shading correction on the current target image by using the difference between the central brightness of the current target image and the brightness of surrounding areas reduced by a pre-calibrated grid;

[0022] Performing automatic white balance processing on the current target image;

[0023] Converting the current target image into a standard color image by performing color interpolation and color matrix correction on the current target image;

[0024] Performing Gamma curve debugging on the current target image;

[0025] By enhancing the high-frequency information of the current target image, edge enhancement of the current target image is achieved;

[0026] Performing color space conversion on the current target image to convert the current target image into an image in a specified format;

[0027] The current target image converted into a specified format is output as the preprocessed current target image.

[0028] Optionally, in the above-mentioned image quality optimization method, the preprocessed current target image is input into a pre-trained image processing model, and the input current target image is processed by the image processing model to obtain multiple optimization parameters of the current target image, including:

[0029] Input the preprocessed current target image into a pre-trained image processing model, and identify feature information of the input current target image in each target dimension through the image processing model; wherein the feature information in each target dimension includes object information, person information and color information;

[0030] Determining various categories of the current target image based on the feature information of the current target image through the image processing model; wherein the categories include scene categories, color cast categories, and color shadow degree categories;

[0031] The optimization parameters of the current target image in color shading correction, white balance correction and color matrix correction are calculated based on the feature information of the current target image and the categories of the current target through the color shading algorithm, white balance algorithm and color restoration algorithm in the image processing model respectively.

[0032] Optionally, in the above-mentioned image quality optimization method, the training method of the image processing model includes:

[0033] Acquire the first sample image and the second sample image collected in multiple scenes;

[0034] Inputting the first sample image and the second sample image into the image processing model;

[0035] The first sample image is convolved by the first convolution network of the image processing model to obtain a first feature map, and the second sample image is convolved by the second convolution network of the image processing model to obtain a second feature map; wherein the first feature map includes various feature information in the first sample image; the second feature map includes various feature information of the second sample image; and the first feature map and the second feature map include the same feature information items;

[0036] The first feature map and the second feature map are sequentially input into multiple algorithm layers, and each algorithm layer is trained sequentially; wherein the multiple algorithm layers include a color shading algorithm layer, a white balance algorithm layer, and a color restoration algorithm layer that are sequentially connected.

[0037] Optionally, in the above-mentioned image quality optimization method, the step of inputting the first feature map and the second feature map into multiple algorithm layers in sequence, and training each algorithm layer in sequence, includes:

[0038] Processing the two input images respectively through the convolution layer in the current algorithm layer to obtain a first convolution image and a second convolution image corresponding to the current algorithm layer; wherein the current algorithm layer refers to the algorithm layer for processing the current input image;

[0039] Determine the category of the first convolution image corresponding to the current algorithm layer based on the first convolution image corresponding to the current algorithm layer through the post-processing layer of the current algorithm layer, and process the first convolution image corresponding to the current algorithm layer and its category using the algorithm corresponding to the current algorithm layer to obtain a current optimization graph corresponding to the current algorithm layer;

[0040] Adjusting the parameters of the algorithm corresponding to the current algorithm layer according to the difference between the current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image in the parameters corresponding to the current algorithm layer;

[0041] The current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image are input into the next algorithm layer, and the above steps are returned to execute until the last algorithm layer, thus ending this round of training.

[0042] The second aspect of the present application provides an image quality optimization system, comprising:

[0043] An image acquisition unit and an image processing and optimization unit; wherein the image processing and optimization unit includes an image processing module and an image quality optimization module;

[0044] The image acquisition unit is used to acquire the current target image captured by the camera;

[0045] The image processing module is used to perform multiple image preprocessing on the current target image;

[0046] The image quality optimization module is used to input the preprocessed current target image into a pre-trained image processing model, process the input current target image through the image processing model, obtain multiple optimization parameters of the current target image, and optimize the current target image using the multiple optimization parameters of the current target image to obtain the optimized current target image; wherein the image processing model identifies the feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on each item of the feature information of the current target image, so as to output the multiple optimization parameters of the current target image based on each item of the feature information of the current target image and each category of the current target image; the image processing model is pre-trained using the first sample image and the second sample image collected in the same scene as input; the first sample image is an image collected when a rearview mirror glass is set in front of the camera; the second sample image is an image collected when no rearview mirror glass is set in front of the camera.

[0047] Optionally, in the above-mentioned image quality optimization system, the image processing and optimization unit is further used for:

[0048] After the system is powered on, the camera is configured to control the camera to start real-time image acquisition and output based on the configured information.

[0049] Optionally, in the above-mentioned image quality optimization system, when the image processing module performs the multiple image preprocessing steps on the current target image, it is used to:

[0050] The current target image is processed using each of the image preprocessing methods in accordance with the processing order of the image preprocessing methods specified by each preset item; wherein each of the image preprocessing methods includes multiple items of automatic exposure control, black level correction, bad pixel correction, RAW noise reduction, lens shading correction, automatic white balance, color interpolation, color matrix correction, gamma curve adjustment, edge enhancement and color space conversion.

[0051] Optionally, in the above-mentioned image quality optimization system, the image processing module executes the processing sequence of the image preprocessing methods specified by the preset items, and when the current target image is processed by using the image preprocessing methods, is used to:

[0052] Calculating the difference between the brightness of the current target image and the preset target brightness, and using the difference to adjust the exposure parameters of the camera;

[0053] Using a preset fixed value, performing black level correction on the current target image;

[0054] Detecting bad pixels in the current target image through gradients in multiple directions, and correcting the bad pixels;

[0055] Filtering out noise information in the current target image by using a filter;

[0056] Performing lens shading correction on the current target image by using the difference between the central brightness of the current target image and the brightness of surrounding areas reduced by a pre-calibrated grid;

[0057] Performing automatic white balance processing on the current target image;

[0058] Converting the current target image into a standard color image by performing color interpolation and color matrix correction on the current target image;

[0059] Performing Gamma curve debugging on the current target image;

[0060] By enhancing the high-frequency information of the current target image, edge enhancement of the current target image is achieved;

[0061] Performing color space conversion on the current target image to convert the current target image into an image in a specified format;

[0062] The current target image converted into a specified format is output as the preprocessed current target image.

[0063] Optionally, in the above-mentioned image quality optimization system, when the image quality optimization module executes the step of inputting the preprocessed current target image into a pre-trained image processing model, processing the input current target image through the image processing model, and obtaining multiple optimization parameters of the current target image, it is used to:

[0064] Input the preprocessed current target image into a pre-trained image processing model, and identify feature information of the input current target image in each target dimension through the image processing model; wherein the feature information in each target dimension includes object information, person information and color information;

[0065] Determining various categories of the current target image based on the feature information of the current target image through the image processing model; wherein the categories include scene categories, color cast categories, and color shadow degree categories;

[0066] The optimization parameters of the current target image in color shading correction, white balance correction and color matrix correction are calculated based on the feature information of the current target image and the categories of the current target through the color shading algorithm, white balance algorithm and color restoration algorithm in the image processing model respectively.

[0067] Optionally, the above-mentioned image quality optimization system further includes a model training unit, wherein the model training unit is used to:

[0068] Acquire the first sample image and the second sample image collected in multiple scenes;

[0069] Inputting the first sample image and the second sample image into the image processing model;

[0070] The first sample image is convolved by the first convolution network of the image processing model to obtain a first feature map, and the second sample image is convolved by the second convolution network of the image processing model to obtain a second feature map; wherein the first feature map includes various feature information in the first sample image; the second feature map includes various feature information of the second sample image; and the first feature map and the second feature map include the same feature information items;

[0071] Inputting the first feature map and the second feature map into a plurality of first algorithm layers in sequence, and training each algorithm layer in sequence; wherein the plurality of algorithm layers include a color shading algorithm layer, a white balance algorithm layer, and a color restoration algorithm layer connected in sequence;

[0072] Optionally, in the above-mentioned image quality optimization system, when the model training unit executes the step of inputting the first feature map and the second feature map into multiple algorithm layers in sequence and training each algorithm layer in sequence, it is used to:

[0073] Processing the two input images respectively through the convolution layer in the current algorithm layer to obtain a first convolution image and a second convolution image corresponding to the current algorithm layer; wherein the current algorithm layer refers to the algorithm layer for processing the current input image;

[0074] Determine the category of the first convolution image corresponding to the current algorithm layer based on the first convolution image corresponding to the current algorithm layer through the post-processing layer of the current algorithm layer, and process the first convolution image corresponding to the current algorithm layer and its category using the algorithm corresponding to the current algorithm layer to obtain a current optimization graph corresponding to the current algorithm layer;

[0075] Adjusting the parameters of the algorithm corresponding to the current algorithm layer according to the difference between the current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image in the parameters corresponding to the current algorithm layer;

[0076] The current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image are input into the next algorithm layer, and the above steps are returned to execute until the last algorithm layer, thus ending this round of training.

[0077] A third aspect of the present application provides an electronic device, including:

[0078] Memory and processor;

[0079] Wherein, the memory is used to store programs;

[0080] The processor is used to execute the program, and when the program is executed, it is specifically used to implement the image quality optimization method as described in any one of the above items.

[0081] A fourth aspect of the present application provides a computer storage medium for storing a computer program, which, when executed, is used to implement the image quality optimization method as described in any one of the above.

[0082] The embodiment of the present application provides an image quality optimization method, which uses the first sample image and the second sample image collected in the same scene as input for training to obtain an image processing model. Among them, the first sample image is an image collected when a rearview mirror glass is set in front of the camera. The second sample image is an image collected when a rearview mirror glass is not set in front of the camera, so it can be guaranteed that the image collected when the rearview mirror glass is set can be optimized to the effect when the rearview mirror glass is not set. In the specific application process, the current target image collected by the camera is obtained, and then multiple image preprocessing is performed on the current target image to first use simple preprocessing to preliminarily optimize the image quality. Then the preprocessed current target image is input into the pre-trained image processing model, and the input current target image is processed by the image processing model to obtain multiple optimization parameters of the current target image. Among them, the image processing model identifies the feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on the various feature information of the current target image, so as to output multiple optimization parameters of the current target image based on the various feature information of the current target image and the various categories of the current target image. Finally, the current target image is optimized using multiple optimization parameters of the current target image to obtain an optimized current target image, thereby optimizing the image to the effect of an image collected when no rearview mirror glass is provided, effectively improving the image quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0084] Figure 1 A flowchart of an image quality optimization method provided in an embodiment of the present application;

[0085] Figure 2 A flowchart of a method for processing a current target image in a processing order provided in an embodiment of the present application;

[0086] Figure 3 A flowchart of a method for obtaining optimization parameters through an image processing model provided in an embodiment of the present application;

[0087] Figure 4 A flowchart of a method for training an image processing model provided in an embodiment of the present application;

[0088] Figure 5A schematic diagram of the architecture of an image processing model provided in an embodiment of the present application;

[0089] Figure 6 A flowchart of a training method for each algorithm layer provided in an embodiment of the present application;

[0090] Figure 7 A schematic diagram of the architecture of an image quality optimization system provided in an embodiment of the present application;

[0091] Figure 8 A schematic diagram of the architecture of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0092] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0093] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0094] The present application embodiment provides a method for optimizing image quality. Figure 1 As shown, including:

[0095] S101, obtaining a current target image captured by a camera.

[0096] Among them, the current target image refers to the image that needs to be optimized for quality during the current acquisition. Specifically, the embodiment of the present application mainly uses the image acquired by the RGB-IR camera installed in the rearview mirror housing as an example for explanation. Therefore, in the embodiment of the present application, the current target image specifically refers to the image that needs to be optimized that is acquired by the current RGB-IR camera through the rearview mirror glass.

[0097] It should be noted that when the camera is an RGB-IR camera, it only receives infrared light with a wavelength of 940nm in IR mode. Therefore, the black and white image will not be affected by the rearview mirror glass of the car, so it does not need to be optimized as the current target image.

[0098] Optionally, in another embodiment of the present application, before executing step S101, the following steps may be further performed:

[0099] After the system is powered on, the camera is configured to control the camera to start real-time image acquisition and output based on the configured information.

[0100] It should be noted that in order to allow the camera to capture images as needed, the camera can be configured after the system is powered on, and after the configuration is completed, the camera starts working, that is, real-time image capture and output.

[0101] S102: Perform multiple image preprocessing steps on the current target image.

[0102] Since the current target image captured through the rearview mirror glass may have some relatively obvious defects, which will affect the accuracy of subsequent fine-grained optimization, and in order to make the image quality higher, in an embodiment of the present application, the current target image will be preprocessed first.

[0103] In addition, corresponding processing can be performed through a variety of different pre-processing methods, such as black level correction, bad pixel correction and noise reduction.

[0104] Optionally, in another embodiment of the present application, a specific implementation of step S102 includes:

[0105] According to the preset processing order of each designated image preprocessing method, the current target image is processed using each image preprocessing method.

[0106] Since various image preprocessing methods may affect each other's image processing, they need to be processed in a certain order.

[0107] Among them, in the embodiment of the present application, each image preprocessing method includes multiple items of automatic exposure control, black level correction, bad pixel correction, RAW noise reduction, lens shading correction, automatic white balance, color interpolation, color matrix correction, Gamma curve adjustment, edge enhancement and color space conversion, and the processing order of each image preprocessing method can be the order of each image preprocessing method listed above. Of course, it can also include the use of other processing orders that will not affect the accuracy of the processing, and other preprocessing methods can also be used to process the current target image.

[0108] Optionally, in another embodiment of the present application, a specific implementation method of processing the current target image according to the processing order is as follows: Figure 2 As shown, the following steps are included:

[0109] S201, calculating the difference between the brightness of the current target image and the preset target brightness, and using the difference to adjust the exposure parameters of the camera.

[0110] In order to ensure that the brightness of the image captured by the camera meets the requirements, in an embodiment of the present application, the exposure parameters of the camera are adjusted in real time based on the difference between the brightness of the currently captured image and the preset target brightness.

[0111] The exposure parameters of the camera may specifically include exposure time and gain.

[0112] S202: Perform black level correction on the current target image using a preset fixed value.

[0113] Specifically, the preset fixed value is subtracted from the pixel value of the current target image, thereby achieving black level correction for the current target image.

[0114] The preset fixed value may be calculated based on an image captured in a pure black scene, or may be set based on an image effect.

[0115] S203, detecting bad pixels in the current target image through gradients in multiple directions, and correcting the bad pixels.

[0116] Optionally, a gradient method may be used to determine bad pixels and perform corrections through gradients in different directions.

[0117] S204: Filter out noise information in the current target image through a filter.

[0118] S205 , performing lens shading correction on the current target image by using the central brightness of the current target image and the difference between the surrounding brightness reduced by the pre-calibrated grid.

[0119] It should be noted that, compared with overall brightness adjustment, lens shading correction is mainly to improve the problem that the periphery of the image is darker than the center due to the lens, so the adjustment is made by using the difference between the center brightness of the current target image and the pre-calibrated grid to reduce the surrounding brightness.

[0120] S206: Perform automatic white balance processing on the current target image.

[0121] Among them, automatic white balance is used to make the image present the correct white under any light source. The specific methods used are usually gray world method or perfect reflection method.

[0122] S207 , converting the current target image into a standard color image by performing color interpolation and color matrix correction on the current target image.

[0123] It should be noted that since the in-car camera directly captures RAW images, they can be converted into RGB images with color information through color difference. Specifically, linear interpolation or color difference and color ratio method can be used. In order to have more standard colors, a coarse-grained correction can also be performed on the color matrix.

[0124] S208, performing Gamma curve debugging on the current target image.

[0125] In order to make the brightness and contrast of the current target image more consistent with the visual characteristics of the human eye, the gamma curve of the current target image can be debugged through linear interpolation or lookup table.

[0126] S209: Enhance the edge of the current target image by enhancing the high-frequency information of the current target image.

[0127] Edge enhancement is used to enhance the details and edges in RGB images, which is generally achieved by enhancing the high-frequency information in the image. Therefore, the edge enhancement of the current target image can be achieved by enhancing the high-frequency information of the current target image.

[0128] S210: Perform color space conversion on the current target image to convert the current target image into an image in a specified format.

[0129] It should be noted that, since images in RAW format with simpler colors are usually directly collected, and the colors need to be analyzed later, it is necessary to convert them into YUV format images with richer colors through color space conversion.

[0130] S211. Output the current target image converted into a specified format as the preprocessed current target image.

[0131] After the above-mentioned processing is performed on the current target image in sequence until it is converted into the specified format, the preprocessing of the current target image is completed, so the current target image converted into the specified format is output as the preprocessed current target image.

[0132] S103, inputting the preprocessed current target image into a pre-trained image processing model, processing the input current target image through the image processing model, and obtaining multiple optimization parameters of the current target image.

[0133] Among them, the image processing model identifies the feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on the feature information of the current target image, so as to output multiple optimization parameters of the current target image based on the feature information of the current target image and the categories of the current target image.

[0134] Since different scenes and situations require different adjustment methods, and the current target image needs to be adjusted in many aspects, it is necessary to extract the feature information of the current target image in multiple target dimensions. Specifically, it may include object information, character information, and color information. Among them, the scene corresponding to the current target image can be judged by the object information and character information, such as the road at noon on a sunny day, the complex traffic at sunset, etc., and the color information can determine whether there is color cast and color shadow, etc. Therefore, the category of the current target image in multiple dimensions can be further determined by the identified feature information, so that targeted analysis can be performed to obtain the corresponding optimization parameters of the current target image in multiple dimensions. Specifically, the corresponding image processing model can be trained according to the defects existing in the collected current target image, so as to obtain the optimization parameters of various defects of the image. Specifically, when analyzing the image collected by the camera behind the rearview mirror glass, it can be the optimization parameters of white balance correction, color matrix correction and color shadow correction.

[0135] The image processing model is trained in advance using the first sample image and the second sample image collected in the same scene as input. The first sample image is an image collected when a rearview mirror glass is set in front of the camera, that is, an image that needs to be optimized. The second sample image is an image collected when no rearview mirror glass is set in front of the camera, that is, an image that does not need to be optimized.

[0136] It should be noted that in the embodiment of the present application, during the model training process, it is necessary to input both the first sample image and the second sample image into the model for processing, so as to directly distinguish the optimization parameters of the first sample image through the model, and compare the optimized parameters with the feature map of the second sample image, so as to accurately adjust the corresponding parameters of the model. There is no need to set up additional units to optimize the first sample using the optimization parameters, compare the processing results of the first sample with the second sample image, and further adjust the parameters of the model. And directly comparing the processing results of the first sample with the second sample image is a comparison of the information of the original image, which is not convenient for accurately determining the parameters that need to be adjusted.

[0137] S104 , optimizing the current target image using multiple optimization parameters of the current target image to obtain an optimized current target image.

[0138] Specifically, after obtaining a plurality of optimization parameters of the current target system in multiple dimensions, corresponding parameters of the current target image may be adjusted based on these optimization parameters, thereby obtaining an optimized current target image.

[0139] Optionally, in another embodiment of the present application, a specific implementation of step S103 is as follows: Figure 3 As shown, the following steps are included:

[0140] S301, input the preprocessed current target image into a pre-trained image processing model, and identify the feature information of the input current target image in each target dimension through the image processing model.

[0141] Among them, the feature information on each target dimension includes object information, person information and color information.

[0142] S302: Determine the categories of the current target image based on feature information of the current target image through an image processing model.

[0143] The categories include scene categories, color cast categories, and color shading degree categories.

[0144] The scene category refers to the scene in which the current target image is captured, such as a road at noon on a sunny day, or complex traffic at sunset, etc. It can be determined based on object information and person information. Since different scenes have certain corresponding characteristics such as brightness, the corresponding scene can be determined to optimize the image in a targeted manner.

[0145] The color cast category refers to whether there is color cast, so as to determine whether optimization is needed for color cast. In addition, the color cast category may also include different levels of existing color cast.

[0146] The color shading degree category is a category corresponding to each degree of color shading in an image, so that the color shading of the image can be corrected according to the color shading degree.

[0147] S303, respectively calculating the optimization parameters of the current target image in color shading correction, white balance correction and color matrix correction through the color shading algorithm, white balance algorithm and color restoration algorithm in the image processing model based on the feature information of the current target image and the categories of the current target.

[0148] Specifically, the color shading algorithm in the image processing model is used to process the information related to color shading in the feature information of the current target image and the categories corresponding to color shading in each category, and the optimization parameters of the current target image in color shading correction are obtained. The white balance algorithm in the image processing model is used to analyze and obtain the optimization parameters of the current target image in white balance correction based on the information related to white balance in the feature information of the current target image and the categories corresponding to white balance in each category. The color restoration algorithm in the image processing model is used to process the information related to color in the feature information of the current target image and the categories corresponding to color in each category, and the optimization parameters of the current target image in color matrix correction are obtained.

[0149] Accordingly, another embodiment of the present application provides a method for training an image processing model, such as Figure 4 As shown, the following steps are included:

[0150] S401: Acquire first sample images and second sample images collected in multiple scenes.

[0151] In order to ensure that the images in each scene can be accurately optimized, in the embodiment of the present application, sample images are collected for multiple scenes. Specifically, for one scene, an image is collected when a rearview mirror glass is set in front of the camera to obtain a first sample image of the scene, and then, when the scene remains unchanged, an image is collected again when no rearview mirror glass is set in front of the camera to obtain a second sample image of the scene.

[0152] S402: Input the first sample image and the second sample image into an image processing model.

[0153] Specifically, Figure 5 As shown, in the trained image processing model, a first convolutional network and a second convolutional network are set to process the first sample image and the second sample image respectively. So specifically, the first sample image is input into the first convolutional network of the image processing model, and the second sample image is input into the second convolutional network of the image processing model.

[0154] It should be noted that after the image processing model is trained, only one input is required in the process of processing the current target image, and the second convolutional network can be removed after training.

[0155] S403, performing convolution processing on the first sample image through the first convolution network of the image processing model to obtain a first feature map, and performing convolution processing on the second sample image through the second convolution network of the image processing model to obtain a second feature map.

[0156] Among them, the first feature map includes various feature information in the first sample image, and the second feature map includes various feature information of the second sample image, that is, in the embodiment of the present application, the identified feature information is presented in the form of a feature map. Therefore, the feature information in the image is identified by the first convolutional network and the second convolutional network. Moreover, the items of feature information included in the first feature map and the second feature map are consistent, so that based on the difference in the processing results of the algorithm layer below the two feature maps, the parameters of the algorithm layer can be adjusted so that the same processing effect can be obtained for the image collected by the rearview mirror glass.

[0157] S404: Input the first feature map and the second feature map into multiple algorithm layers in sequence, and train each algorithm layer in sequence.

[0158] The multiple algorithm layers include a color shading algorithm layer, a white balance algorithm layer, and a color restoration algorithm layer that are connected in sequence. Figure 5 As shown, in the embodiment of the present application, a corresponding algorithm layer is set for each algorithm, and each algorithm layer is connected in sequence. Therefore, it is necessary to process through each algorithm layer in sequence, so it is necessary to first input the first algorithm layer for processing, and then adjust its parameters according to the difference in the processing results of the two input feature maps, so as to realize a training of the first algorithm layer. Then the two output results of the first algorithm layer, that is, the two processed feature images output, are input to the next algorithm layer for processing. Similarly, based on the processing results of the next layer, its parameters are adjusted to realize a training of the next algorithm layer, and then its output results are output to the next layer, so that each algorithm layer is continuously trained until the training of the last algorithm layer is completed, and an iterative training is completed. Of course, since the processing process of each algorithm is independent, even if it is interconnected, the processing of the image by the upper algorithm layer will not affect the processing of the next algorithm layer, so each algorithm layer can also be disconnected.

[0159] And, if Figure 5 As shown in the figure, each algorithm layer includes a convolution layer and a post-processing layer. The convolution layer is used to extract the feature graph containing the feature information required for the algorithm analysis process of the algorithm layer, while the post-processing layer is mainly used to determine the category based on the feature information and train the algorithm based on the feature information and category information. Figure 5 As shown, the color shading algorithm layer includes a color shading convolution layer and a color shading post-processing layer, the white balance algorithm layer includes a white balance convolution layer and a white balance post-processing layer, and the color restoration algorithm layer includes a color restoration convolution layer and a color restoration post-processing layer. Therefore, the processing flow of each algorithm layer is the same, only the algorithms and information used are different.

[0160] Optionally, in another embodiment of the present application, a specific implementation of step S404 is as follows: Figure 6 As shown, including:

[0161] S601. Process two input images respectively through the convolution layer in the current algorithm layer to obtain a first convolution image and a second convolution image corresponding to the current algorithm layer.

[0162] The current algorithm layer refers to the algorithm layer for processing the current input image, that is, initially it is the first algorithm layer, and each time step S605 is executed, it changes to the next algorithm layer.

[0163] It should be noted that for the first algorithm layer, the two input images are the outputs of the first convolutional network and the second convolutional network, and for the subsequent algorithm layers, namely the second algorithm layer and the third algorithm layer, the two input images are the output images of the previous algorithm layer, that is, the corresponding current optimization image and the corresponding second convolutional image output by the previous algorithm layer.

[0164] Specifically, the current algorithm layer performs convolution processing on the two input images through its convolution layer, so as to obtain a feature map containing the required feature information.

[0165] S602. Determine the category of the first convolution image corresponding to the current algorithm layer based on the first convolution image corresponding to the current algorithm layer through the post-processing layer of the current algorithm layer, and use the algorithm corresponding to the current algorithm layer to process the first convolution image corresponding to the current algorithm layer and its category to obtain a current optimization graph corresponding to the current algorithm layer.

[0166] Specifically, by analyzing the first convolution image and its category corresponding to the current algorithm layer through the algorithm corresponding to the current algorithm layer, the optimization parameters corresponding to the current algorithm can be obtained, and the first convolution image corresponding to the current algorithm can be optimized using the parameters, so that the current optimization graph corresponding to the current algorithm layer can be obtained.

[0167] It should be noted that since there is no need to perform image comparison or adjust the algorithm parameters after training, in actual applications, when using this image processing model, the post-processing layer can only obtain the optimization parameters corresponding to the algorithm, and there is no need to optimize the corresponding convolutional map.

[0168] S603: Adjust the parameters of the algorithm corresponding to the current algorithm layer by comparing the difference between the current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image in the parameters corresponding to the current algorithm layer.

[0169] It should be noted that the parameters corresponding to the color shading algorithm layer are color shading parameters, the parameters corresponding to the white balance algorithm layer are parameters corresponding to white balance, and the parameters corresponding to the color restoration algorithm layer are color matrix parameters.

[0170] Specifically, if the difference between the current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image in the parameters corresponding to the algorithm layer has not converged, the parameters of the algorithm are adjusted accordingly according to the difference. If the difference between the two has converged, no further adjustment is required.

[0171] S604: Determine whether the current algorithm layer belongs to the last layer.

[0172] If the current algorithm layer is not the last layer, then step S605 is executed. If the current algorithm layer is the last layer, then step S606 is executed.

[0173] S605: Input the current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image into the next algorithm layer.

[0174] Since the processing information of each algorithm layer is different and independent of each other, the image processed by the previous layer can be directly output to the next layer for processing. Figure 5 As shown, the feature maps that have not been processed multiple times are re-input.

[0175] It should be noted that in the embodiment of the present application, when a round of iterative training is performed, each algorithm is trained accordingly. Therefore, after the previous algorithm layer is processed, it is processed by the next algorithm layer. Therefore, after executing step S606, it is necessary to return to execute the above steps, that is, return to execute again from step S601. Of course, it is also possible to perform multiple rounds of iterative training on the previous algorithm layer until the parameters of the corresponding algorithm are determined, and then transfer to the next algorithm layer for processing, that is, after the training of the previous algorithm is completed, the next algorithm is trained.

[0176] S606: End this round of training.

[0177] Specifically, after a round of training, if there are still untrained algorithms among the three algorithms, it is necessary to start the next round of iterative training. If the final parameters of the three algorithms have been determined, the training of the image processing model can be calculated.

[0178] Optionally, in another embodiment of the present application, after executing step S104, the following may be further included:

[0179] The optimized current target image is displayed on the monitor.

[0180] The embodiment of the present application provides an image quality optimization method, which uses the first sample image and the second sample image collected in the same scene as input for training to obtain an image processing model. Among them, the first sample image is an image collected when a rearview mirror glass is set in front of the camera. The second sample image is an image collected when a rearview mirror glass is not set in front of the camera, so it can be guaranteed that the image collected when the rearview mirror glass is set can be optimized to the effect when the rearview mirror glass is not set. In the specific application process, the current target image collected by the camera is obtained, and then multiple image preprocessing is performed on the current target image to first use simple preprocessing to preliminarily optimize the image quality. Then the preprocessed current target image is input into the pre-trained image processing model, and the input current target image is processed by the image processing model to obtain multiple optimization parameters of the current target image. Among them, the image processing model identifies the feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on the various feature information of the current target image, so as to output multiple optimization parameters of the current target image based on the various feature information of the current target image and the various categories of the current target image. Finally, the current target image is optimized using multiple optimization parameters of the current target image to obtain an optimized current target image, thereby optimizing the image to the effect of an image collected when no rearview mirror glass is provided, effectively improving the image quality.

[0181] Another embodiment of the present application provides an image quality optimization system, such as Figure 7 As shown, including:

[0182] Image acquisition unit 701 and image processing and optimization unit 702.

[0183] Among them, Figure 7 As shown, the image processing and optimization unit includes an image processing module 7021 and an image quality optimization module 7022.

[0184] Optionally, the image acquisition unit 701 may include a camera disposed behind the rearview mirror. The image acquisition unit 701 may send the collected image to the image processing and optimization unit 702 via serial MIPI communication. The image processing and optimization unit 702 may transmit the camera in the image acquisition unit via IIC communication to send a control signal, configure it, and start the controller to work.

[0185] The image acquisition unit 701 is used to acquire the current target image captured by the camera.

[0186] The image processing module 7021 is used to perform multiple image preprocessing operations on the current target image.

[0187] The image quality optimization module 7022 is used to input the pre-processed current target image into a pre-trained image processing model, process the input current target image through the image processing model, obtain multiple optimization parameters of the current target image, and optimize the current target image using the multiple optimization parameters of the current target image to obtain the optimized current target image. Among them, the image processing model identifies the feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on the various feature information of the current target image, so as to output multiple optimization parameters of the current target image based on the various feature information of the current target image and the various categories of the current target image. The image processing model is pre-trained using the first sample image and the second sample image collected in the same scene as input. The first sample image is an image collected when a rearview mirror glass is set in front of the camera. The second sample image is an image collected when no rearview mirror glass is set in front of the camera.

[0188] Optionally, in the image quality optimization system provided in another embodiment of the present application, the image processing and optimization unit is further used to:

[0189] After the system is powered on, the camera is configured to control the camera to start real-time image acquisition and output based on the configured information.

[0190] Optionally, in the image quality optimization system provided in another embodiment of the present application, when the image processing module performs multiple image preprocessing operations on the current target image, it is used to:

[0191] According to the preset processing order of each designated image preprocessing method, the current target image is processed using each image preprocessing method.

[0192] Among them, the various image preprocessing methods include automatic exposure control, black level correction, bad pixel correction, RAW noise reduction, lens shading correction, automatic white balance, color interpolation, color matrix correction, gamma curve adjustment, edge enhancement and color space conversion.

[0193] Optionally, in an image quality optimization system provided in another embodiment of the present application, the image processing module executes the processing sequence of each preset specified image preprocessing method, and when processing the current target image using each image preprocessing method, it is used to:

[0194] Calculate the difference between the brightness of the current target image and the preset target brightness, and use the difference to adjust the exposure parameters of the camera.

[0195] Use the preset fixed value to perform black level correction on the current target image.

[0196] The bad pixels in the current target image are detected through gradients in multiple directions and corrected.

[0197] The noise information in the current target image is filtered out through a filter.

[0198] The lens shading correction is performed on the current target image by utilizing the difference between the central brightness of the current target image and the brightness of the surrounding areas reduced by a pre-calibrated grid.

[0199] Perform automatic white balance processing on the current target image.

[0200] The current target image is converted into a standard color image by performing color interpolation and color matrix correction on the current target image.

[0201] Debug the gamma curve of the current target image.

[0202] By enhancing the high-frequency information of the current target image, the edge enhancement of the current target image is achieved.

[0203] Performs color space conversion on the current target image to convert the current target image into an image of the specified format.

[0204] The current target image converted to the specified format is output as the preprocessed current target image.

[0205] Optionally, in the above-mentioned image quality optimization system, when the image quality optimization module inputs the preprocessed current target image into a pre-trained image processing model, processes the input current target image through the image processing model, and obtains multiple optimization parameters of the current target image, it is used to:

[0206] The preprocessed current target image is input into a pre-trained image processing model, and the image processing model is used to identify the feature information of the input current target image in each target dimension, wherein the feature information in each target dimension includes object information, person information and color information.

[0207] The image processing model is used to determine various categories of the current target image based on the feature information of the current target image, wherein the categories include scene categories, color cast categories, and color shadow degree categories.

[0208] The color shading algorithm, white balance algorithm and color restoration algorithm in the image processing model are used respectively to calculate the optimization parameters of the current target image in color shading correction, white balance correction and color matrix correction based on the feature information of the current target image and the various categories of the current target.

[0209] Optionally, in another embodiment of the present application, the image quality optimization system further includes a model training unit, the model training unit is used to:

[0210] A first sample image and a second sample image collected in multiple scenes are obtained.

[0211] The first sample image and the second sample image are input into the image processing model.

[0212] The first sample image is convolved by a first convolutional network of the image processing model to obtain a first feature map, and the second sample image is convolved by a second convolutional network of the image processing model to obtain a second feature map.

[0213] The first feature map includes various feature information in the first sample image. The second feature map includes various feature information in the second sample image. The first feature map and the second feature map include the same feature information.

[0214] The first feature map and the second feature map are sequentially input into multiple algorithm layers, and each algorithm layer is trained at one time.

[0215] The multiple algorithm layers include a color shading algorithm layer, a white balance algorithm layer and a color restoration algorithm layer which are connected in sequence.

[0216] Optionally, in the image quality optimization system provided in another embodiment of the present application, when the model training unit executes the step of inputting the first feature map and the second feature map into multiple algorithm layers in sequence and training each algorithm layer in sequence, it is used to:

[0217] The first feature map and the second feature map are input into the first algorithm layer.

[0218] The two input images are processed respectively by the convolution layer in the current algorithm layer to obtain the first convolution image and the second convolution image corresponding to the current algorithm layer.

[0219] The current algorithm layer refers to the algorithm layer for processing the current input image.

[0220] Through the post-processing layer of the current algorithm layer, based on the first convolution image corresponding to the current algorithm layer, the category of the first convolution image corresponding to the current algorithm layer is determined, and the first convolution image corresponding to the current algorithm layer and its category are processed by the algorithm corresponding to the current algorithm layer to obtain the current optimization graph corresponding to the current algorithm layer.

[0221] The parameters of the algorithm corresponding to the current algorithm layer are adjusted by the difference between the current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image in the parameters corresponding to the algorithm layer.

[0222] The current optimization graph and the corresponding second convolution image corresponding to the current algorithm layer are input into the next algorithm layer, and the two input images are processed respectively through the convolution layer in the current algorithm layer to obtain the first convolution image and the second convolution image corresponding to the current algorithm layer, until the last algorithm layer, ending this round of training.

[0223] Optionally, in the image quality optimization system provided in another embodiment of the present application, also refer to Figure 7 , also includes:

[0224] The image display unit 703 is used to display the optimized current target image on the display.

[0225] Alternatively, if Figure 7 As shown, the image display unit 703 may specifically include a display.

[0226] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the implementation process of the corresponding steps in the above method embodiments, and will not be repeated here.

[0227] Another embodiment of the present application provides an electronic device, such as Figure 8 As shown, including:

[0228] Memory 801 and processor 802 .

[0229] The memory 801 is used to store programs.

[0230] The processor 802 is used to execute the program stored in the memory 801, and when the program is executed, it is specifically used to implement the image quality optimization method provided in any one of the above embodiments.

[0231] Another embodiment of the present application provides a computer storage medium for storing a computer program. When the computer program is executed by a processor, it is used to implement the image quality optimization method provided in any of the above embodiments.

[0232] Computer storage media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0233] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0234] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing image quality, characterized in that: include: Get the current target image captured by the camera; Performing multiple image preprocessing on the current target image; The preprocessed current target image is input into a pre-trained image processing model, and the input current target image is processed by the image processing model to obtain multiple optimization parameters of the current target image; wherein the image processing model identifies feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on each item of feature information of the current target image, so as to output multiple optimization parameters of the current target image based on each item of feature information of the current target image and each category of the current target image; the image processing model is trained in advance using a first sample image and a second sample image collected in the same scene as input; the first sample image is an image collected when a rearview mirror glass is set in front of the camera; the second sample image is an image collected when no rearview mirror glass is set in front of the camera; The current target image is optimized using a plurality of optimization parameters of the current target image to obtain the optimized current target image.

2. The method according to claim 1, characterized in that Before obtaining the current target image captured by the camera, the method further includes: After the system is powered on, the camera is configured to control the camera to start real-time image acquisition and output based on the configured information.

3. The method according to claim 1, characterized in that The performing multiple image preprocessing on the current target image includes: The current target image is processed using each of the image preprocessing methods in accordance with the processing order of the image preprocessing methods specified by each preset item; wherein each of the image preprocessing methods includes multiple items of automatic exposure control, black level correction, bad pixel correction, RAW noise reduction, lens shading correction, automatic white balance, color interpolation, color matrix correction, gamma curve adjustment, edge enhancement and color space conversion.

4. The method according to claim 3, characterized in that The processing of the current target image by using each of the image preprocessing methods according to the preset processing order of each of the image preprocessing methods includes: Calculating the difference between the brightness of the current target image and the preset target brightness, and using the difference to adjust the exposure parameters of the camera; Using a preset fixed value, performing black level correction on the current target image; Detecting bad pixels in the current target image through gradients in multiple directions, and correcting the bad pixels; Filtering out noise information in the current target image by using a filter; Performing lens shading correction on the current target image by using the difference between the central brightness of the current target image and the brightness of surrounding areas reduced by a pre-calibrated grid; Performing automatic white balance processing on the current target image; Converting the current target image into a standard color image by performing color interpolation and color matrix correction on the current target image; Performing Gamma curve debugging on the current target image; By enhancing the high-frequency information of the current target image, edge enhancement of the current target image is achieved; Performing color space conversion on the current target image to convert the current target image into an image in a specified format; The current target image converted into a specified format is output as the preprocessed current target image.

5. The method according to claim 1, characterized in that: The preprocessed current target image is input into a pre-trained image processing model, and the input current target image is processed by the image processing model to obtain multiple optimization parameters of the current target image, including: Input the preprocessed current target image into a pre-trained image processing model, and identify feature information of the input current target image in each target dimension through the image processing model; wherein the feature information in each target dimension includes object information, person information and color information; Determining various categories of the current target image based on the feature information of the current target image through the image processing model; wherein the categories include scene categories, color cast categories, and color shadow degree categories; The optimization parameters of the current target image in color shading correction, white balance correction and color matrix correction are calculated based on the feature information of the current target image and the categories of the current target through the color shading algorithm, white balance algorithm and color restoration algorithm in the image processing model respectively.

6. The method according to claim 1, characterized in that The training method of the image processing model comprises: Acquire the first sample image and the second sample image collected in multiple scenes; Inputting the first sample image and the second sample image into the image processing model; The first sample image is convolved by the first convolution network of the image processing model to obtain a first feature map, and the second sample image is convolved by the second convolution network of the image processing model to obtain a second feature map; wherein the first feature map includes various feature information in the first sample image; the second feature map includes various feature information of the second sample image; and the first feature map and the second feature map include the same feature information items; The first feature map and the second feature map are sequentially input into multiple algorithm layers, and each algorithm layer is trained sequentially; wherein the multiple algorithm layers include a color shading algorithm layer, a white balance algorithm layer, and a color restoration algorithm layer that are sequentially connected.

7. The method according to claim 6, characterized in that The step of sequentially inputting the first feature map and the second feature map into a plurality of algorithm layers, and sequentially training each algorithm layer, comprises: Processing the two input images respectively through the convolution layer in the current algorithm layer to obtain a first convolution image and a second convolution image corresponding to the current algorithm layer; wherein the current algorithm layer refers to the algorithm layer for processing the current input image; Determine the category of the first convolution image corresponding to the current algorithm layer based on the first convolution image corresponding to the current algorithm layer through the post-processing layer of the current algorithm layer, and process the first convolution image corresponding to the current algorithm layer and its category using the algorithm corresponding to the current algorithm layer to obtain a current optimization graph corresponding to the current algorithm layer; Adjusting the parameters of the algorithm corresponding to the current algorithm layer according to the difference between the current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image in the parameters corresponding to the current algorithm layer; The current optimization graph corresponding to the current algorithm layer and the corresponding second convolution image are input into the next algorithm layer, and the above steps are returned to execute until the last algorithm layer, thus ending this round of training.

8. An image quality optimization system, characterized in that: include: An image acquisition unit and an image processing and optimization unit; wherein the image processing and optimization unit includes an image processing module and an image quality optimization module; The image acquisition unit is used to acquire the current target image captured by the camera; The image processing module is used to perform multiple image preprocessing on the current target image; The image quality optimization module is used to input the preprocessed current target image into a pre-trained image processing model, process the input current target image through the image processing model, obtain multiple optimization parameters of the current target image, and optimize the current target image using the multiple optimization parameters of the current target image to obtain the optimized current target image; wherein the image processing model identifies the feature information of the current target image in each target dimension, and determines the category of the current target image in multiple dimensions based on each item of the feature information of the current target image, so as to output the multiple optimization parameters of the current target image based on each item of the feature information of the current target image and each category of the current target image; the image processing model is pre-trained using the first sample image and the second sample image collected in the same scene as input; the first sample image is an image collected when a rearview mirror glass is set in front of the camera; the second sample image is an image collected when no rearview mirror glass is set in front of the camera.

9. An electronic device, characterized in that: include: Memory and processor; Wherein, the memory is used to store programs; The processor is used to execute the program, and when the program is executed, it is specifically used to implement the image quality optimization method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: Used to store a computer program, which, when executed, is used to implement the image quality optimization method as described in any one of claims 1 to 7.

Images