Evolution of image processing models: methods, devices, media, and equipment

Abstract

This application provides a method for evolving an image processing model, an apparatus for evolving an image processing model, a computer-readable storage medium, and an electronic device, relating to the field of computer technology. The method can train the current image processing model by combining the target image set corresponding to the current time period and the historical image set corresponding to the historical time period, thereby improving the accuracy of the trained model.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a method for evolving an image processing model, an apparatus for evolving an image processing model, a computer-readable storage medium, and an electronic device. Background Technology

[0002] With advancements in image processors and central processing units (CPUs), the efficiency and effectiveness of image processing models have significantly improved. Generally, models need to learn and evolve through a certain number of samples before processing images to enhance their image processing accuracy. In related technologies, a common approach is to first collect and label a certain amount of image data to obtain sample images, which then trigger the model to learn from these sample images. However, models evolved in this way often suffer from low accuracy.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute related technology known to those skilled in the art. Summary of the Invention

[0004] The purpose of this application is to provide an image processing model evolution method, an image processing model evolution device, a computer-readable storage medium, and an electronic device, which can train the current image processing model by combining the target image set corresponding to the current time period and the historical image set corresponding to the historical time period, thereby improving the accuracy of the trained model.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0006] According to one aspect of this application, an evolution method for an image processing model is provided, the method comprising:

[0007] Get the set of target images corresponding to the current time period;

[0008] Obtain the set of historical images corresponding to the historical time period;

[0009] The current image processing model is trained based on the target image set and the historical image set to obtain the target image processing model.

[0010] According to one aspect of this application, an evolution apparatus for an image processing model is provided, the apparatus comprising:

[0011] The image acquisition unit is used to acquire the set of target images corresponding to the current time period;

[0012] The image acquisition unit is also used to acquire a set of historical images corresponding to a historical time period;

[0013] The model evolution unit is used to train the current image processing model based on the target image set and the historical image set to obtain the target image processing model.

[0014] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.

[0015] According to one aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the method of any one of the above.

[0016] According to one aspect of this application, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method of any of the above by executing the executable instructions.

[0017] The exemplary embodiments of this application may have some or all of the following beneficial effects:

[0018] In an example embodiment of this application, the image processing model evolution method can be used to train the current image processing model by combining the target image set corresponding to the current time period and the historical image set corresponding to the historical time period, thereby improving the accuracy of the trained model.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0021] Figure 1 A flowchart illustrating an evolution method of an image processing model according to an embodiment of this application is shown schematically.

[0022] Figure 2 A schematic diagram of a client interface according to an embodiment of this application is shown.

[0023] Figure 3This illustration schematically shows an image filtering process according to one embodiment of the present application;

[0024] Figure 4 A schematic diagram illustrating the results of multiple cameras according to an embodiment of this application is shown.

[0025] Figure 5 This illustration schematically shows an evolutionary strategy diagram according to an embodiment of the present application;

[0026] Figure 6 A flowchart illustrating an evolution method of an image processing model according to another embodiment of this application is shown schematically;

[0027] Figure 7 The diagram schematically illustrates a system architecture diagram of an evolutionary method for applying an image processing model according to an embodiment of this application;

[0028] Figure 8 This schematic diagram illustrates the structure of an evolution apparatus for an image processing model according to an embodiment of the present application;

[0029] Figure 9 The schematic diagram illustrates the structure of a computer system suitable for implementing the electronic devices of the present application. Detailed Implementation

[0030] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of the embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this application.

[0031] Please see Figure 1 , Figure 1 A flowchart illustrating the evolution method of an image processing model according to an embodiment of this application is shown schematically. Figure 1 As shown, the method includes the following steps.

[0032] Step S110: Obtain the set of target images corresponding to the current time period.

[0033] Step S120: Obtain the set of historical images corresponding to the historical time period.

[0034] Step S130: Train the current image processing model based on the target image set and the historical image set to obtain the target image processing model.

[0035] Implementation Figure 1 The method shown can combine the target image set corresponding to the current time period and the historical image set corresponding to the historical time period to train the current image processing model, thereby improving the accuracy of the trained model.

[0036] The steps described above in this example implementation will now be explained in more detail.

[0037] As an optional embodiment, the method further includes: triggering the client to display an image optimization function control; and after receiving a request from the client to enable the image optimization function control, acquiring an image set according to a preset period. This enhances interactivity with the user. Based on the user-enabled image optimization function, an image set can be acquired periodically. The image set may include images uploaded by clients of one or more users who have enabled the image optimization function. This facilitates model evolution through the periodically acquired image set, and the evolved model can further optimize the images uploaded by the client, achieving a positive evolution effect.

[0038] For details, please refer to Figure 2 , Figure 2 A schematic diagram of a client interface according to an embodiment of this application is shown. Figure 2 As shown, users connected to a specific network can intuitively understand the control 210 for activating NAS storage and the control 220 for joining the automatic portrait effect upgrade plan (i.e., the image optimization function control mentioned above) through a client installed on a terminal device (e.g., a mobile phone or tablet). By triggering controls 210 and 220, the image optimization function can be enabled simultaneously with NAS storage activation. The image optimization function is used to trigger the image processing model to process the acquired images (e.g., blurring, contrast optimization, etc.). Furthermore, in response to the trigger operation of control 220 (i.e., the image optimization function control mentioned above), the client can generate an activation request and send it to the NAS device / cloud server. After receiving the activation request, the NAS device / cloud server can obtain an image set according to a preset period (e.g., one day, one week, one month, etc.).

[0039] In this context, NAS (Network Attached Storage) refers to network-attached storage. On hardware devices with computing capabilities, the NAS device can function as a local storage device. One or more users can connect to the NAS device via a specific network and transfer data (such as images, text, videos, and other multimedia files) stored on their terminal devices to the NAS device. This solves the problem of limited storage space on user-owned terminal devices. Users can upload data to the NAS device when their terminal device's storage space is insufficient, or they can upload newly generated data to the NAS device. After data is uploaded to the NAS device, users can choose to delete the same data stored on their terminal device or not. When the data is needed, it can be retrieved from the NAS device and then read back onto the terminal device. Within the NAS device, data uploaded by different users can be isolated from each other (i.e., different users can only access their own uploaded data from the NAS device, and cannot access data uploaded by other users), or data sharing can be achieved (i.e., users can access data uploaded by other users from the NAS device). This embodiment of the application does not limit this.

[0040] In step S110, the set of target images corresponding to the current time period is obtained.

[0041] Specifically, the target image set corresponding to the current time period (e.g., 2022-11-01 to 2022-11-06) can be understood as the latest target image set obtained according to a preset period, where the acquisition time of each target image in the target image set is within the latest preset period (e.g., 2022-11-01 to 2022-11-06). Furthermore, the target images in the target image set can come from different clients or from the same client; this embodiment does not limit this. The number of images in the target image set obtained within different preset periods can be different, depending on the upload volume of each client within that preset period.

[0042] As an optional embodiment, obtaining the target image set corresponding to the current time period includes: obtaining the image set corresponding to the current time period; and filtering the image set to obtain the target image set. This filtering of the image set can prevent non-compliant images from entering the image processing flow and consuming computational resources, thus saving computational resources.

[0043] Specifically, the number of images corresponding to the target image set obtained after filtering can be less than or equal to the image set. If all images are determined to be compliant during the filtering process, all images in the image set can be retained without being filtered.

[0044] As an optional embodiment, obtaining the image set corresponding to the current time period includes: determining the images uploaded by each client through a specific network within the current time period as the image set. This enables centralized processing of images corresponding to each client under a specific network, and model evolution based on diverse images uploaded by different clients can help improve the accuracy of model evolution.

[0045] Specifically, the specific network can be either a private network or a public network, and this application does not limit it in its embodiments.

[0046] As an optional embodiment, the image set is filtered to obtain a target image set, including: images that do not meet a preset brightness range, images that do not meet a preset focus distance range, images that do not contain human subjects, images that have shooting occlusion, images that are duplicates of other images in the image set, images that have a pixel position consistency parameter less than a first threshold among multiple cameras, and images that have a pixel luminance consistency parameter less than a second threshold among multiple cameras, to obtain the target image set. This allows for the filtering of one or more of the above seven types of images according to different needs, thereby obtaining a target image set that can participate in model evolution, which is beneficial for improving the accuracy of model evolution.

[0047] Specifically, the number of images in categories one, two, three, four, five, six, and seven can be the same or different, and this application embodiment does not limit this. The detection of images in categories one, two, three, four, five, six, and seven can be performed in parallel, sequentially in the order of categories one, two, three, four, five, six, and seven, or only one or more types of images can be detected; this application embodiment does not limit this.

[0048] It should be noted that if the detection of images of type 1, type 2, type 3, type 4, type 5, type 6, and type 7 is performed sequentially, then the detection of type 2 images is based on the image set after type 1 images have been filtered out, and so on, with continuous detection of type 3, type 4, type 5, type 6, and type 7 images.

[0049] Based on this, the above embodiment can be specifically implemented as follows: First, a first reference image set is obtained by filtering out images from the image set that do not meet a preset brightness range; second, a second reference image set is obtained by filtering out images from the first reference image set that do not meet a preset focus distance range; third, a third reference image set is obtained by filtering out images from the second reference image set that do not contain human subjects; fourth, a fourth reference image set is obtained by filtering out images from the third reference image set that exhibit occlusion during shooting; fifth, a fifth reference image set is obtained by filtering out images from the fourth reference image set that repeat with other images in the image set; sixth, a sixth reference image set is obtained by filtering out images from the fifth reference image set that have a pixel position consistency parameter between multiple cameras that is less than a first threshold; and finally, a target image set is obtained by filtering out at least one of seven types of images from the sixth reference image set that have a pixel luminance consistency parameter between multiple cameras that is less than a second threshold.

[0050] For details regarding the above embodiments, please refer to [link / reference]. Figure 3 , Figure 3 A schematic diagram illustrating an image filtering process according to an embodiment of this application is shown. Figure 3 As shown, the image set 310 can be sequentially filtered out in the following order: images 311 that do not meet the preset brightness range, images 312 that do not meet the preset focus distance range, images 313 that do not contain human subjects, images 314 that have shooting occlusion, images 315 that are duplicates of other images in the image set, images 316 that have pixel position consistency parameters between multiple cameras that are less than the first threshold, and images 317 that have pixel luminance consistency parameters between multiple cameras that are less than the second threshold, thereby obtaining the target image set 320.

[0051] As an optional embodiment, the method further includes: performing image quality detection on each image in the image set to obtain a brightness value corresponding to each image; and identifying images whose brightness values ​​are outside a preset brightness range (e.g., 50–100) as a class of images. This can prevent images that are generally too bright or too dark from participating in the model evolution, which is beneficial to improving the accuracy of the model evolution.

[0052] The preset brightness range can be limited by an upper limit brightness value and a lower limit brightness value. The upper limit brightness value can be used to limit the maximum brightness value of the image, and the lower limit brightness value can be used to limit the minimum brightness value of the image. Images with brightness values ​​greater than the upper limit brightness value or less than the lower limit brightness value belong to the same category.

[0053] Specifically, performing image quality detection on each image in the image set to obtain the corresponding brightness value for each image can be implemented as follows: detecting the brightness value corresponding to each image in the image set, where the brightness value is used to characterize the overall brightness of the image, to obtain the brightness value corresponding to each image. Optionally, the brightness value corresponding to each image in the image set can be detected by: detecting the brightness value corresponding to each pixel in the image, calculating the average of the brightness values ​​of each pixel to obtain the brightness value of the corresponding image. In this way, the brightness value corresponding to each image can be calculated.

[0054] As an optional embodiment, the method further includes: reading the metadata of each image in the image set; obtaining the focus distance from the metadata; and classifying images whose focus distance is outside a preset focus distance range as Class II images. This avoids images with excessively close / far focus distances from participating in model evolution, which helps improve the accuracy of model evolution.

[0055] The preset focus distance range can be limited by an upper focus distance and a lower focus distance. The upper focus distance can be used to limit the farthest focus distance of the image, and the lower focus distance can be used to limit the closest focus distance of the image. Images that are greater than the upper focus distance or less than the lower focus distance belong to the second category.

[0056] In addition, each image in the uploaded image set corresponds to different metadata, which is used to record image shooting parameters, such as focus distance and shooting results of each camera.

[0057] As an optional embodiment, the method further includes: performing human detection on each image in the image set to obtain human detection results corresponding to each image; selecting target human detection results that do not contain human subjects; and determining the images corresponding to the target human detection results as three types of images. This avoids images that do not contain human subjects from participating in model evolution, which is beneficial for improving the blurring effect of human figures after model evolution.

[0058] Specifically, human detection is performed on each image in the image set to obtain human detection results corresponding to each image. This includes: performing feature detection on each image in the image set based on preset feature vectors (e.g., eye feature vector, nose feature vector, body feature vector, etc.) to obtain human detection results corresponding to each image. Furthermore, the human detection results used to represent targets that do not contain human subjects can be represented in any form, such as text, vectors, or charts; this application embodiment does not impose any limitations.

[0059] In addition, optionally, it may include: performing plant / animal detection on each image in the image set to obtain the plant / animal detection result corresponding to each image; selecting the target plant / animal detection result to characterize the plant / animal that does not contain the plant / animal; and identifying the image corresponding to the target plant / animal detection result as eight types of images and filtering them out.

[0060] As an optional embodiment, the method further includes: acquiring the multi-camera results corresponding to each image in the image set; calculating the histogram difference value to characterize the differences between the multi-camera results based on the histograms corresponding to the capture results of each camera in the multi-camera results, so as to obtain the histogram difference value corresponding to each image; and identifying the images corresponding to the histogram difference value greater than a preset value as four types of images. This can avoid extreme images with occlusion participating in the model evolution, which is beneficial to improving the accuracy of model evolution.

[0061] Specifically, the results captured by each camera in the multi-camera setup may include the results captured by the main camera and the results captured by the secondary cameras. The terminal device may include multiple secondary cameras. Therefore, the results captured by the secondary cameras may be one or more, which is not limited in the embodiments of this application.

[0062] Furthermore, the histograms corresponding to the capture results of each camera in the multi-camera results can be used to characterize the brightness information, color information, etc., corresponding to the capture results of the respective cameras. Based on this, according to the histograms corresponding to the capture results of each camera in the multi-camera results, a histogram difference value is calculated to characterize the differences between the multi-camera results. This includes: calculating the histogram difference value between the capture result of each secondary camera and the capture result of the main camera based on the histograms corresponding to the capture results of each camera in the multi-camera results, so as to obtain the histogram difference value used to characterize the differences between the multi-camera results.

[0063] It should be noted that if an image corresponds to multiple histogram difference values, then each of the multiple histogram difference values ​​needs to be compared with a preset value. If any histogram difference value is greater than the preset value, then the image is classified as a Class IV image.

[0064] Alternatively, the histogram difference between the shooting results of each secondary camera and the shooting results of the main camera can be calculated by subtracting the histogram of the secondary camera from the histogram of the main camera to obtain the histogram difference between the shooting results of the secondary camera and the shooting results of the main camera. Based on this, the histogram difference between the shooting results of each secondary camera and the shooting results of the main camera can be obtained.

[0065] As an optional embodiment, the method further includes: acquiring the multi-camera results corresponding to each image in the image set; determining the correlation between each image based on the main camera result in the multi-camera results; and for each group of images with a correlation higher than a preset correlation, identifying any image within that group as a Class 5 image. This avoids duplicate images from participating in the model evolution, which is beneficial for improving the accuracy of the model evolution.

[0066] Specifically, the correlation between images is determined based on the main camera result in the multi-camera results, including: calculating the hash value of the main camera result between each image, and calculating the correlation between each pair of hash values. If the correlation is higher than the preset correlation, it indicates that the two images may be duplicate images, and then either of the two images can be identified as a five-category image.

[0067] As an optional embodiment, the method further includes: acquiring multi-camera results corresponding to each image in the image set; wherein the multi-camera results include main camera results and secondary camera results; generating pixel correspondence between the main camera results and secondary camera results for each multi-camera result; generating pixel position consistency parameters corresponding to each image based on the pixel correspondence; and identifying images with pixel position consistency parameters less than a first threshold (e.g., 50%) as six types of images. This can avoid extreme image model evolution with low pixel position consistency among multi-camera results, which is beneficial to improving the accuracy of model evolution.

[0068] Specifically, it can generate pixel correspondence between the main camera result and each secondary camera result. The pixel correspondence is used to characterize the positional correspondence of the same pixel in the main camera result and the secondary camera result.

[0069] Taking a dual-camera setup as an example (i.e., an image contains a main camera result and a secondary camera result), for each multi-camera result, the pixel correspondence between the main camera result and the secondary camera result is generated. This includes: determining the shooting posture and the hardware positional relationship between the main camera and the secondary camera based on the image's metadata. If the shooting posture is vertical and the hardware positional relationship between the main camera and the secondary camera is horizontal, or if the shooting posture is horizontal and the hardware positional relationship between the main camera and the secondary camera is vertical, then the pixels in the main camera result and the secondary camera result correspond to the same Y coordinate and different X coordinates. This can be determined based on the expression... The formulas X_R=X_L-k / f(X_L,Y_L),Y_R=Y_L and X_L=k / f(X_R,Y_R)+X_R,Y_R=Y_L determine the pixel correspondence between each pixel in the main camera result and the secondary camera result; where X_R and X_L are used to represent the different X coordinates of the same pixel in the main camera result and the secondary camera result, respectively, Y_R and Y_L are used to represent the same Y coordinate of the same pixel in the main camera result and the secondary camera result, respectively, k is used to characterize the multi-camera system constant for depth-of-field to parallax conversion, and f is used to characterize the mapping relationship between coordinates.

[0070] Furthermore, if the shooting posture is horizontal and the hardware positions of the main camera and the secondary camera are arranged horizontally, or if the shooting posture is vertical and the hardware positions of the main camera and the secondary camera are arranged vertically, then the pixels in the main camera result and the secondary camera result correspond to the same X coordinate and different Y coordinates. The pixel correspondence between each pixel in the main camera result and the secondary camera result can be determined according to the expressions X_L=X_R-k / f(X_R,Y_R),Y_L=Y_R and X_R=k / f(X_L,Y_L)+X_L,Y_L=Y_R. Here, X_R and X_L are used to represent the same X coordinate of the same pixel in the main camera result and the secondary camera result, respectively, and Y_R and Y_L are used to represent the different Y coordinates of the same pixel in the main camera result and the secondary camera result, respectively.

[0071] For details, please refer to [link / reference]. Figure 4 , Figure 4 A schematic diagram illustrating the results of multiple cameras according to an embodiment of this application is shown. Figure 4 As shown, if the shooting posture is vertical and the hardware positions of the main camera and the secondary camera are arranged horizontally, or if the shooting posture is horizontal and the hardware positions of the main camera and the secondary camera are arranged vertically, then the pixel 411 at the center of the shadow area in the main camera result 410 and the pixel 421 at the center of the shadow area in the secondary camera result 420 correspond to the same Y coordinate and different X coordinates, that is, Y_R and Y_L are the same, while X_R and X_L are different.

[0072] As an optional embodiment, a pixel position consistency parameter corresponding to each image is generated based on the pixel correspondence relationship between each image. This includes: generating a set of pixel difference values ​​corresponding to each image based on the pixel correspondence relationship between each image; wherein each set of pixel difference values ​​includes the position difference value corresponding to each pixel in the corresponding image; for each set of pixel difference values, calculating the proportion of pixel difference values ​​less than a pixel difference value threshold; and determining the proportion of pixel difference values ​​as the pixel position consistency parameter for the corresponding image. This allows for the determination of the proportion of pixel difference values ​​less than the pixel difference value threshold, thereby facilitating the accurate determination of pixel position consistency.

[0073] Specifically, each pixel in an image corresponds to a pixel correspondence. For each pixel, a pixel difference value (diff) can be generated based on the expression `diff = |X_L – X_R|`. For an image, the pixel difference values ​​(diff) of each pixel can form a set of pixel difference values ​​corresponding to that image, and each image can correspond to a set of pixel difference values. Within a set of pixel difference values, the number of pixel difference values ​​(diff) less than a pixel difference value threshold (e.g., 0.5) can be calculated, and the ratio of this number to the total number of diff values ​​in the set is determined as the percentage of pixel difference values ​​corresponding to the set (e.g., 80%). The lower the percentage of pixel difference values, the lower the pixel position consistency in the image corresponding to that set. Such images are not suitable for model evolution and can be classified as Class 6 images. The pixel difference value threshold can be set to a constant.

[0074] As an optional embodiment, the method further includes: generating a set of luminance difference values ​​corresponding to each image based on the pixel correspondence of each image; wherein each set of luminance difference values ​​includes the luminance difference values ​​corresponding to each pixel in the corresponding image; generating a luminance consistency parameter corresponding to each image based on the set of luminance difference values ​​corresponding to each image; and identifying images with a pixel luminance consistency parameter less than a second threshold (e.g., 60%) as seven types of images. This can avoid extreme image model evolution with low pixel luminance consistency among multi-camera results, which is beneficial to improving the accuracy of model evolution.

[0075] Specifically, based on the pixel correspondence of each image, a set of luminance difference values ​​corresponding to each image is generated. This includes: calculating the luminance difference value diff_illum corresponding to each pixel in the image based on the pixel correspondence of each image and the expression diff_illum=|IL(X_L,Y_L)–IR(X_R,Y_R)|, thereby obtaining a set of luminance difference values ​​for each image. Performing the above steps for each image yields the set of luminance difference values ​​for each image. Here, IL and IR represent the main camera result and the secondary camera result, respectively.

[0076] In addition, based on the set of photometric difference values ​​corresponding to each image, a photometric consistency parameter corresponding to each image is generated, including: for each set of photometric difference values, calculating the proportion of photometric difference values ​​less than the photometric difference value threshold (e.g., 90%); determining the proportion of photometric difference values ​​as the photometric location consistency parameter of the corresponding image; wherein, the lower the proportion of photometric difference values, the lower the photometric consistency in the image corresponding to the set of photometric difference values, and such images are not suitable for participating in model evolution and can be classified as seven types of images.

[0077] In step S120, the set of historical images corresponding to the historical time period is obtained.

[0078] Specifically, a historical period can be understood as a time period defined by a predetermined historical cycle.

[0079] As an optional embodiment, obtaining the historical image set corresponding to the historical time period includes: obtaining the historical image sets corresponding to multiple consecutive historical time periods adjacent to the current time period. This allows multiple historical image sets to be incorporated into the model evolution process, thereby improving the model's evolution accuracy.

[0080] Specifically, if the current time period is represented as T, then multiple consecutive historical time periods can be represented as T-1, T-2, T-3, etc. In this application, the specific number of multiple historical time periods is not limited. In addition, the historical image sets corresponding to multiple historical time periods can contain different numbers of images.

[0081] Understandably, as time progresses and the next preset cycle arrives, the current time period becomes T+1, while T, T-1, T-2, T-3, etc., become historical time periods.

[0082] In step S130, the current image processing model is trained based on the target image set and the historical image set to obtain the target image processing model.

[0083] Specifically, the current image processing model can be any image processing architecture, and it can belong to a machine learning model or other model type that can automatically learn features.

[0084] As an optional embodiment, the method further includes: blurring each target image in the target image set using a target image processing model, and storing the blurred target images. This allows users to easily obtain more refined blurred target images when they need to retrieve them again, avoiding processing target images on the user's client side, not consuming client computing resources, and reducing the user's perception of the image optimization process.

[0085] Specifically, the target image in the target image set is blurred through the target image processing model, and the blurred target image is stored. This includes: blurring each target image through the layer architecture of the target image processing model, and the blurred target image can be stored in the storage unit of the NAS device so that users can access it at any time.

[0086] As an optional embodiment, the method further includes: sending the blurred target image to the client in response to the client's target image reading request. This allows for timely response to the user's image acquisition request, providing the user with a more refined target image.

[0087] Optionally, in response to a client's target image read request, sending the blurred target image to the client includes: authenticating the target image read request in response to the client's target image read request; if the authentication result indicates that the client is legitimate, then sending the blurred target image to the client. This can prevent unauthorized users from stealing data from legitimate users through the client.

[0088] As an optional embodiment, the current image processing model is trained based on the target image set and the historical image set to obtain the target image processing model. This includes: obtaining the first pixel correspondence of each target image in the target image set according to the current image processing model corresponding to the current time period; obtaining the second pixel correspondence of each target image according to the original image processing model; updating the target image set according to the first pixel correspondence and the second pixel correspondence of each target image; and training the current image processing model based on the updated target image set and the historical image set to obtain the target image processing model. This allows for model evolution based on the original image processing model and the current image processing model, avoiding the situation where model training deviates from the initial target due to using only the first pixel correspondence of each target image in the target image set obtained by the current image processing model for model training, thus improving the accuracy of model evolution.

[0089] Specifically, the first pixel correspondence and the second pixel correspondence can be different, as they are obtained based on image processing models at different time periods. The first pixel correspondence can be understood as the pixel correspondence of the target image already obtained; it is named "first pixel correspondence" here to distinguish it from the second pixel correspondences of each target image obtained from the original image processing model. Furthermore, the original image processing model M0 can be the initial, untrained image processing model.

[0090] As an optional embodiment, updating the target image set based on the first pixel correspondence and the second pixel correspondence of each target image includes: determining the union of the first pixel correspondence and the second pixel correspondence of each target image to obtain the union of relations corresponding to each target image; generating sparsity corresponding to each union of relations to characterize the pixel defect rate; determining the number of sparsities greater than a preset threshold; if the number is less than a preset number, resampling the target images corresponding to sparsities greater than the preset threshold to obtain resampled images; and adding the resampled images to the target image set. This improves the image richness of the target image set by resampling the target images corresponding to sparsities greater than the preset threshold, thereby improving the accuracy of model evolution.

[0091] Specifically, for each image, the union of the first pixel correspondence and the second pixel correspondence is calculated to obtain the union of relations for each target image. Then, based on the number of pixel difference values ​​greater than or equal to a pixel difference threshold and the number of photometric difference values ​​greater than or equal to a photometric difference threshold in each union of relations, sparsity is generated for each union of relations to characterize the pixel defect rate. The sum of the number of photometric difference values ​​and the number of pixel difference values ​​divided by the total number of pixels represents the sparsity, thus obtaining the sparsity corresponding to each target image. Furthermore, the sparsity of each target image can be sorted, and the number of sparsity values ​​greater than a preset threshold is determined based on the sorting result. If the number of sparsity values ​​is less than a preset number (e.g., 100), the target image corresponding to the sparsity values ​​greater than the preset threshold is resampled (i.e., copied), resulting in a resampled image (i.e., the copied image). The resampled image can then be added to the target image set to improve the image richness in the target image set.

[0092] Please see Figure 5 , Figure 5 A schematic diagram illustrating an evolutionary strategy according to an embodiment of this application is shown. Figure 5 As shown, on the timeline, based on a preset period, each historical period can correspond to a different image set, namely, the target image set 511 corresponding to T, the historical image set 512 corresponding to T-1, the historical image set 513 corresponding to T-2, and the historical image set 514 corresponding to T-3. Furthermore, different periods can also correspond to different image processing models, namely, the current image processing model 521 and the historical image processing model 522. For the current period, the current image processing model 521 needs to be evolved. This can be achieved by combining the target image set 511 corresponding to T, the historical image set 512 corresponding to T-1, the historical image set 513 corresponding to T-2, the historical image set 514 corresponding to T-3, the preset sample image set 531, and the historical image processing model 522, thereby improving the evolution accuracy of the current image processing model 521. It should be noted that after the next preset period arrives, the current image processing model 521 will become a historical image processing model.

[0093] As an optional embodiment, the current image processing model is trained based on the updated target image set and the historical image set to obtain the target image processing model. This includes: training the current image processing model based on the updated target image set, the historical image set, and the preset sample image set to obtain the target image processing model. This allows for model evolution using data from three dimensions: the updated target image set, the historical image set, and the preset sample image set, which helps improve the model's evolution accuracy.

[0094] Specifically, the preset sample image set can be a pre-labeled image set, and the number of sample images included in the preset sample image set is not limited in this embodiment. The current image processing model is trained based on the updated target image set, historical image set, and preset sample image set to obtain the target image processing model. This includes: the current image processing model jointly learning image features from the target image set, historical image set, and preset sample image set to obtain the target image processing model.

[0095] As an optional embodiment, the method further includes: obtaining the historical image processing model weights corresponding to the previous time period; and updating the weights of the current image processing model based on the historical image processing model weights. This enables the updating of the current image processing model weights based on the historical image processing model weights, thereby improving the model's evolution accuracy.

[0096] The weights corresponding to the current image processing model in the current time period can be represented as W. T The weights of the historical image processing model corresponding to the previous time period can be represented as W. T-1 In W T and W T-1 The model may include one or more weight parameters, which are not limited in this embodiment. Specifically, updating the weights corresponding to the current image processing model based on the weights of the historical image processing model includes: adjusting the weights W corresponding to the current image processing model. T Weights W of the historical image processing model T-1 Substitute into expression W T =W T-1 +(1–β)*W T The moving average inertia fusion process is performed to obtain the final target image processing model M after weight update. T β can be represented as a constant.

[0097] Please see Figure 6 , Figure 6 A flowchart illustrating the evolution of an image processing model according to another embodiment of this application is shown schematically. Figure 6 As shown, the evolution method of this image processing model includes steps S610 to S642.

[0098] Step S610: Trigger the client to display the image optimization function control. After receiving the client's request to enable the image optimization function control, obtain the image set according to the preset period.

[0099] Step S612: Determine the images uploaded by each client through a specific network during the current time period as an image set.

[0100] Step S614: Perform image quality detection on each image in the image set to obtain the brightness value corresponding to each image, and identify images whose brightness values ​​are outside the preset brightness range as a type of image.

[0101] Step S616: Read the metadata of each image in the image set, obtain the focus distance in the metadata, and identify images whose focus distance is outside the preset focus distance range as Class II images.

[0102] Step S618: Perform human detection on each image in the image set to obtain the human detection result corresponding to each image, and select the target human detection result to represent the target human subject that does not contain a human subject, and determine the image corresponding to the target human detection result as three types of images.

[0103] Step S620: Obtain the multi-camera results corresponding to each image in the image set. Based on the histograms corresponding to the capture results of each camera in the multi-camera results, calculate the histogram difference value used to characterize the differences between the multi-camera results, so as to obtain the histogram difference value corresponding to each image. The images corresponding to the histogram difference value that is greater than the preset value are identified as four types of images.

[0104] Step S622: Obtain the multi-camera results corresponding to each image in the image set, determine the correlation between each image based on the main camera result in the multi-camera results, and for each group of images with a correlation higher than the preset correlation, determine any image in the group as a five-category image.

[0105] Step S624: Obtain the multi-camera results corresponding to each image in the image set; wherein, the multi-camera results include the main camera results and the secondary camera results; for each multi-camera result, generate the pixel correspondence between the main camera results and the secondary camera results; based on the pixel correspondence corresponding to each image, generate a set of pixel difference values ​​corresponding to each image; wherein, each set of pixel difference values ​​includes the positional difference values ​​corresponding to each pixel in the corresponding image; for each set of pixel difference values, calculate the proportion of pixel difference values ​​less than the pixel difference value threshold; determine the proportion of pixel difference values ​​as the pixel position consistency parameter of the corresponding image; determine the images whose pixel position consistency parameter is less than the first threshold as six types of images.

[0106] Step S626: Generate a set of photometric difference values ​​corresponding to each image based on the pixel correspondence of each image; wherein each set of photometric difference values ​​includes the photometric difference value corresponding to each pixel in the corresponding image; generate a photometric consistency parameter corresponding to each image based on the set of photometric difference values ​​corresponding to each image; and identify images whose pixel photometric consistency parameter is less than the second threshold as seven types of images.

[0107] Step S628: Filter out at least one of the following from the image set: images that do not meet the preset brightness range, images that do not meet the preset focus distance range, images that do not contain human subjects, images that have shooting occlusion, images that are duplicates of other images in the image set, images that have a pixel position consistency parameter between multiple cameras less than a first threshold, and images that have a pixel luminance consistency parameter between multiple cameras less than a second threshold, to obtain the target image set.

[0108] Step S630: Obtain the set of historical images corresponding to multiple consecutive historical time periods adjacent to the current time period.

[0109] Step S632: Based on the current image processing model corresponding to the current time period, obtain the first pixel correspondence of each target image in the target image set, and obtain the second pixel correspondence of each target image based on the original image processing model.

[0110] Step S634: Determine the union of the first pixel correspondence and the second pixel correspondence of each target image to obtain the union of the relations corresponding to each target image, and generate the sparsity corresponding to each union of relations to characterize the pixel defect rate. Then determine the number of sparsities greater than a preset threshold. If the number is less than the preset number, resample the target images corresponding to the sparsity greater than the preset threshold to obtain resampled images, and add the resampled images to the target image set.

[0111] Step S636: Train the current image processing model based on the updated target image set and the historical image set to obtain the target image processing model.

[0112] Step S638: Obtain the weights of the historical image processing model corresponding to the previous time period of the current time period, and update the weights of the current image processing model according to the weights of the historical image processing model.

[0113] Step S640: Blur each target image in the target image set using the target image processing model, and store the blurred target images.

[0114] Step S642: In response to the client's target image reading request, send the blurred target image to the client.

[0115] It should be noted that steps S610 to S642 are related to... Figure 1 For the specific implementation details of steps S610 to S642, please refer to the examples shown. Figure 1 The steps and their embodiments shown are not repeated here.

[0116] It is evident that implementation Figure 6The method described can train the current image processing model by combining the target image set corresponding to the current time period and the historical image set corresponding to historical time periods, thereby improving the accuracy of the trained model. Furthermore, it can enhance user interactivity. Based on the image optimization function enabled by the user, an image set can be periodically acquired. This image set can include images uploaded by one or more users whose clients have enabled image optimization. This facilitates model evolution through periodically acquired image sets, and the evolved model can further optimize images uploaded by the clients, achieving a positive evolution effect. In addition, by filtering the image set, non-compliant images can be prevented from entering the image processing flow and consuming computational resources, thus saving computational resources.

[0117] Please see Figure 7 , Figure 7 This schematically illustrates a system architecture diagram of an image processing model evolution method applicable to an embodiment of this application. The image processing model evolution method disclosed in this application can be applied to... Figure 7 The system shown.

[0118] Different clients can correspond to different users. Figure 7 The clients 730, 740, and 750 shown are merely examples. In actual applications, the number of clients connected to a specific network 720 can be any number, and this application embodiment does not limit this.

[0119] Specifically, clients 730, 740, and 750 can upload images to the NAS device / cloud server 710 through a specific network 720. The NAS device / cloud server 710 can acquire image sets according to a preset period. For each image acquisition, the following steps can be performed: acquire the target image set corresponding to the current time period, and acquire the historical image sets corresponding to multiple consecutive historical time periods adjacent to the current time period.

[0120] Then, the calculation unit 711 can filter out at least one of the following from the image set: images that do not meet the preset brightness range, images that do not meet the preset focus distance range, images that do not contain human subjects, images that have shooting occlusion, images that are duplicates of other images in the image set, images that have pixel position consistency parameters between multiple cameras less than a first threshold, and images that have pixel luminance consistency parameters between multiple cameras less than a second threshold, so as to obtain the target image set.

[0121] Furthermore, based on the current image processing model corresponding to the current time period, the first pixel correspondence of each target image in the target image set is obtained, and the second pixel correspondence of each target image is obtained based on the original image processing model. Additionally, the union of the first and second pixel correspondences of each target image is determined to obtain the union of relations corresponding to each target image, and a sparsity representing the pixel defect rate is generated for each union of relations. The number of sparsities greater than a preset threshold is then determined. If the number is less than a preset number, the target images corresponding to sparsities greater than the preset threshold are resampled to obtain resampled images, which are added to the target image set. The current image processing model is then trained based on the updated target image set and the historical image set to obtain the target image processing model. Finally, the weights of the historical image processing model corresponding to the previous time period are obtained, and the weights of the current image processing model are updated based on the historical image processing model weights.

[0122] Furthermore, the target images in the target image set are blurred using a target image processing model, and the blurred target images are stored in storage unit 712. Subsequently, upon receiving a target image read request, in response to a target image read request from client 730, client 740, or client 750, the blurred target images can be sent to client 730, client 740, or client 750.

[0123] The computing unit 711 may include components such as a central processing unit (CPU), a neural-network processing unit (NPU), and a graphics processing unit (GPU), but this embodiment does not limit the scope. Furthermore, the specific network 720 may be a public network or a private network, but this embodiment does not limit the scope.

[0124] Please see Figure 8 , Figure 8 A schematic block diagram illustrating the evolution of an image processing model according to one embodiment of this application is shown. Figure 8 As shown, the image processing model evolution device 800 may include the following units.

[0125] Image acquisition unit 801 is used to acquire a set of target images corresponding to the current time period;

[0126] The image acquisition unit 801 is also used to acquire a set of historical images corresponding to the historical time period;

[0127] The model evolution unit 802 is used to train the current image processing model based on the target image set and the historical image set to obtain the target image processing model.

[0128] It is evident that implementation Figure 8 The device shown can train the current image processing model by combining the target image set corresponding to the current time period and the historical image set corresponding to the historical time period, thereby improving the accuracy of the trained model.

[0129] As an optional embodiment, it also includes:

[0130] The trigger unit is used to trigger the client to display the image optimization function control;

[0131] The image acquisition unit 801 is also used to acquire an image set according to a preset period after receiving a request from the client to enable the image optimization function control.

[0132] As can be seen, implementing this optional embodiment can improve interactivity with users. Based on the image optimization function enabled by the user, an image set can be periodically acquired. The image set may include images uploaded by one or more users' clients who have enabled the image optimization function. This facilitates model evolution through the periodically acquired image set. The evolved model can further optimize the images uploaded by the client, achieving a positive evolution effect.

[0133] As an optional embodiment, the image acquisition unit 801 acquires a set of target images corresponding to the current time period, including:

[0134] Get the image set corresponding to the current time period;

[0135] The image set is filtered to obtain the target image set.

[0136] As can be seen, by implementing this optional embodiment, non-compliant images can be prevented from entering the image processing flow and consuming computing resources by filtering the image set, thus saving computing resources.

[0137] As an optional embodiment, the image acquisition unit 801 acquires the image set corresponding to the current time period, including:

[0138] The images uploaded by each client through a specific network during the current time period are identified as an image set.

[0139] As can be seen, implementing this optional embodiment can achieve centralized processing of images corresponding to each client under a specific network, and model evolution based on diverse images uploaded by different clients can help improve the accuracy of model evolution.

[0140] As an optional embodiment, the image acquisition unit 801 filters the image set to obtain a target image set, including:

[0141] The target image set is obtained by filtering out at least one of the following from the image set: images that do not meet the preset brightness range, images that do not meet the preset focus distance range, images that do not contain human subjects, images that have shooting occlusion, images that are duplicates of other images in the image set, images that have a pixel position consistency parameter between multiple cameras less than a first threshold, and images that have a pixel luminance consistency parameter between multiple cameras less than a second threshold.

[0142] As can be seen, by implementing this optional embodiment, one or more of the above seven types of images can be filtered out according to different needs, thereby obtaining a set of target images that can participate in model evolution, which is beneficial to improving the accuracy of model evolution.

[0143] As an optional embodiment, it also includes:

[0144] An image detection unit is used to perform image quality detection on each image in the image set to obtain the brightness value corresponding to each image; images whose brightness values ​​are outside the preset brightness range are identified as a class of images.

[0145] As can be seen, implementing this optional embodiment can avoid the overall overly bright / overly dark image from participating in the model evolution, which is beneficial to improving the accuracy of the model evolution.

[0146] As an optional embodiment, it also includes:

[0147] The second-class image detection unit is used to read the metadata of each image in the image set; obtain the focus distance in the metadata; and identify images whose focus distance is outside the preset focus distance range as second-class images.

[0148] As can be seen, implementing this optional embodiment can prevent images with excessively close / far focus distances from participating in model evolution, which is beneficial for improving the accuracy of model evolution.

[0149] As an optional embodiment, it also includes:

[0150] The three image detection units are used to perform human detection on each image in the image set to obtain human detection results corresponding to each image; select target human detection results that do not contain human subjects; and determine the images corresponding to the target human detection results as three types of images.

[0151] As can be seen, implementing this optional embodiment can avoid the participation of images that do not contain human subjects in the model evolution, which is beneficial to improving the blurring effect of human figures after the model evolution.

[0152] As an optional embodiment, it also includes:

[0153] Four image detection units are used to acquire the multi-camera results corresponding to each image in the image set; based on the histograms corresponding to the capture results of each camera in the multi-camera results, the histogram difference value used to characterize the differences between the multi-camera results is calculated to obtain the histogram difference value corresponding to each image; the images corresponding to the histogram difference value greater than the preset value are identified as the four types of images.

[0154] As can be seen, implementing this optional embodiment can avoid extreme images with occlusion conditions from participating in model evolution, which is beneficial to improving the accuracy of model evolution.

[0155] As an optional embodiment, it also includes:

[0156] Five image detection units are used to acquire the multi-camera results corresponding to each image in the image set; determine the correlation between each image based on the main camera result in the multi-camera results; and for each group of images with a correlation higher than the preset correlation, any image in the group is identified as a five-category image.

[0157] As can be seen, implementing this optional embodiment can avoid duplicate images from participating in model evolution, which is beneficial to improving the accuracy of model evolution.

[0158] As an optional embodiment, it also includes:

[0159] Six image detection units are used to acquire multi-camera results corresponding to each image in the image set; wherein the multi-camera results include main camera results and secondary camera results; for each multi-camera result, a pixel correspondence relationship between the main camera result and the secondary camera result is generated; based on the pixel correspondence relationship corresponding to each image, a pixel position consistency parameter corresponding to each image is generated; images with a pixel position consistency parameter less than a first threshold are identified as six types of images.

[0160] As can be seen, implementing this optional embodiment can avoid extreme image model evolution with low pixel position consistency among multi-camera results, which is beneficial to improving the accuracy of model evolution.

[0161] As an optional embodiment, the six image detection units generate pixel position consistency parameters corresponding to each image based on the pixel correspondence between each image, including:

[0162] Based on the pixel correspondence of each image, a set of pixel difference values ​​corresponding to each image is generated; wherein, each set of pixel difference values ​​includes the positional difference values ​​of each pixel in the corresponding image;

[0163] For each set of pixel difference values, calculate the percentage of pixel difference values ​​that are less than the pixel difference value threshold.

[0164] The percentage of pixel differences is determined as the pixel position consistency parameter for the corresponding image.

[0165] As can be seen, implementing this optional embodiment can reduce the proportion of pixel differences that are less than the pixel difference threshold, thereby facilitating the accurate determination of pixel position consistency.

[0166] As an optional embodiment, it also includes:

[0167] The seven-category image detection unit is used to generate a set of photometric difference values ​​corresponding to each image based on the pixel correspondence of each image; wherein each set of photometric difference values ​​includes the photometric difference value corresponding to each pixel in the corresponding image; and to generate a photometric consistency parameter corresponding to each image based on the set of photometric difference values ​​corresponding to each image; and to determine images whose pixel photometric consistency parameter is less than a second threshold as seven-category images.

[0168] As can be seen, implementing this optional embodiment can avoid extreme image model evolution with low pixel luminance consistency among multi-camera results, which is beneficial to improving the accuracy of model evolution.

[0169] As an optional embodiment, the image acquisition unit 801 acquires a set of historical images corresponding to a historical time period, including:

[0170] Obtain the set of historical images corresponding to multiple consecutive historical time periods adjacent to the current time period.

[0171] As can be seen, implementing this optional embodiment allows multiple historical image sets to be incorporated into the model evolution process, thereby improving the model's evolution accuracy.

[0172] As an optional embodiment, it also includes:

[0173] The image blurring unit is used to blur each target image in the target image set through the target image processing model and to store the blurred target image.

[0174] As can be seen, implementing this optional embodiment can provide users with a more refined, blurred target image when they acquire the target image again, avoiding processing the target image on the user's client, not occupying the client's computing resources, and reducing the user's perception of the image optimization process.

[0175] As an optional embodiment, it also includes:

[0176] The response unit is used to respond to the client's target image reading request and send the blurred target image to the client.

[0177] As can be seen, implementing this optional embodiment can respond to users' image acquisition requests in a timely manner and provide users with more refined target images.

[0178] As an optional embodiment, the model evolution unit 802 trains the current image processing model based on the target image set and the historical image set to obtain the target image processing model, including:

[0179] Based on the current image processing model corresponding to the current time period, obtain the first pixel correspondence of each target image in the target image set;

[0180] The second pixel correspondence of each target image is obtained based on the original image processing model;

[0181] Update the target image set according to the first pixel correspondence and the second pixel correspondence of each target image;

[0182] The current image processing model is trained based on the updated target image set and the historical image set to obtain the target image processing model.

[0183] As can be seen, implementing this optional embodiment can realize model evolution based on the original image processing model and the current image processing model, avoiding the situation where the model training deviates from the initial target by only using the first pixel correspondence of each target image in the target image set obtained by the current image processing model for model training, which is beneficial to improving the accuracy of model evolution.

[0184] As an optional embodiment, the model evolution unit 802 updates the target image set according to the first pixel correspondence and the second pixel correspondence of each target image, including:

[0185] Determine the union of the first pixel correspondence and the second pixel correspondence of each target image to obtain the union of the corresponding relationships for each target image;

[0186] Generate the sparsity corresponding to the union of each relation to characterize the pixel defect rate;

[0187] Determine the number of sparsities greater than a preset threshold. If the number is less than the preset number, resample the target image corresponding to the sparsity greater than the preset threshold to obtain a resampled image.

[0188] Add the resampled image to the target image set.

[0189] As can be seen, by implementing this optional embodiment, the image richness of the target image set can be improved by resampling the target images corresponding to sparsity greater than a preset threshold, which is beneficial to improving the model evolution accuracy.

[0190] As an optional embodiment, the model evolution unit 802 trains the current image processing model based on the updated target image set and the historical image set to obtain the target image processing model, including:

[0191] The current image processing model is trained based on the updated target image set, historical image set, and preset sample image set to obtain the target image processing model.

[0192] As can be seen, implementing this optional embodiment allows for model evolution using data from three dimensions: the updated target image set, the historical image set, and the preset sample image set, which helps improve the model's evolution accuracy.

[0193] As an optional embodiment, it also includes:

[0194] The weight update unit is used to obtain the historical image processing model weights corresponding to the previous time period of the current time period; and to update the weights corresponding to the current image processing model based on the historical image processing model weights.

[0195] As can be seen, implementing this optional embodiment can achieve the updating of the current image processing model weights based on the historical image processing model weights, thereby helping to improve the evolution accuracy of the model.

[0196] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0197] Since the functional modules of the image processing model evolution apparatus in the example embodiments of this application correspond to the steps of the image processing model evolution apparatus in the example embodiments described above, for details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the image processing model evolution apparatus described above.

[0198] Please see Figure 9 , Figure 9 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0199] It should be noted that, Figure 9 The computer system 900 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0200] like Figure 9As shown, the computer system 900 includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 902 or programs loaded from storage section 908 into random access memory (RAM) 903. The RAM 903 also stores various programs and data required for system operation. The CPU 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0201] The following components are connected to I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to I / O interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 910 as needed so that computer programs read from it can be installed into storage section 908 as needed.

[0202] In particular, according to embodiments of this application, the processes described below with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by central processing unit (CPU) 901, it performs the various functions defined in the methods and apparatus of this application.

[0203] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to implement the methods described in the above embodiments.

[0204] It should be noted that the computer-readable medium shown in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0205] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0206] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0207] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

Claims

1. An evolution method for an image processing model, characterized in that, include: Get the set of target images corresponding to the current time period; Obtain the set of historical images corresponding to the historical time period; The current image processing model is trained based on the target image set and the historical image set to obtain the target image processing model, including: Based on the current image processing model corresponding to the current time period, obtain the first pixel correspondence relationship of each target image in the target image set; The second pixel correspondence of each target image is obtained based on the original image processing model; The target image set is updated according to the first pixel correspondence and the second pixel correspondence of each target image; The current image processing model is trained based on the updated target image set and the historical image set to obtain the target image processing model.

2. The method according to claim 1, characterized in that, Also includes: Trigger the client to display the image optimization function control; After receiving the client's request to enable the image optimization function control, the image set is obtained according to a preset period.

3. The method according to claim 1, characterized in that, Also includes: The target image processing model is used to blur each target image in the target image set, and the blurred target images are stored.

4. The method according to claim 3, characterized in that, Also includes: In response to the client's target image reading request, the blurred target image is sent to the client.

5. The method according to claim 1, characterized in that, Obtain the set of target images corresponding to the current time period, including: Get the image set corresponding to the current time period; The image set is filtered to obtain the target image set.

6. The method according to claim 5, characterized in that, Retrieve the image set corresponding to the current time period, including: The images uploaded by each client through a specific network during the current time period are identified as an image set.

7. The method according to claim 5, characterized in that, The image set is filtered to obtain a target image set, including: The target image set is obtained by filtering out at least one of the following from the image set: images that do not meet the preset brightness range, images that do not meet the preset focus distance range, images that do not contain human subjects, images that have shooting occlusion, images that are duplicates of other images in the image set, images that have a pixel position consistency parameter between multiple cameras less than a first threshold, and images that have a pixel luminance consistency parameter between multiple cameras less than a second threshold.

8. The method according to claim 7, characterized in that, Also includes: Image quality detection is performed on each image in the image set to obtain the brightness value corresponding to each image; Images whose brightness values ​​are outside the preset brightness range are identified as the first type of image.

9. The method according to claim 7, characterized in that, Also includes: Read the metadata of each image in the image set; Obtain the focus distance from the metadata; Images whose focus distance is outside the preset focus distance range are identified as the second type of image.

10. The method according to claim 7, characterized in that, Also includes: Human detection is performed on each image in the image set to obtain human detection results corresponding to each image; Select the target human detection results that do not contain human subjects; The images corresponding to the target human body detection results are identified as the three types of images.

11. The method according to claim 7, characterized in that, Also includes: Obtain the multi-camera results corresponding to each image in the image set; Based on the histograms corresponding to the capture results of each camera in the multi-camera results, calculate the histogram difference value to characterize the differences between the multi-camera results, so as to obtain the histogram difference value corresponding to each image respectively; Images corresponding to histogram differences greater than a preset value are identified as the four types of images.

12. The method according to claim 7, characterized in that, Also includes: Obtain the multi-camera results corresponding to each image in the image set; The correlation between the images is determined based on the primary camera result in the multi-camera results. For each group of images with a relevance higher than a preset relevance, any image within that group is identified as one of the five image categories.

13. The method according to claim 7, characterized in that, Also includes: Obtain the multi-camera results corresponding to each image in the image set; wherein, the multi-camera results include the main camera results and the secondary camera results; For each multi-camera result, generate the pixel correspondence between the main camera result and the secondary camera result; Based on the pixel correspondence of each image, a pixel position consistency parameter corresponding to each image is generated; Images with a pixel position consistency parameter less than the first threshold are identified as the six types of images.

14. The method according to claim 13, characterized in that, Based on the pixel correspondence of each image, a pixel position consistency parameter corresponding to each image is generated, including: Based on the pixel correspondence of each image, a set of pixel difference values ​​corresponding to each image is generated; wherein, each set of pixel difference values ​​includes the positional difference values ​​of each pixel in the corresponding image; For each set of pixel difference values, calculate the percentage of pixel difference values ​​that are less than the pixel difference value threshold. The percentage of pixel differences is determined as the pixel position consistency parameter of the corresponding image.

15. The method according to claim 13, characterized in that, Also includes: Based on the pixel correspondence of each image, a set of photometric difference values ​​corresponding to each image is generated; wherein, each set of photometric difference values ​​includes the photometric difference value corresponding to each pixel in the corresponding image; Based on the set of photometric difference values ​​corresponding to each image, a photometric consistency parameter corresponding to each image is generated. Images whose pixel luminance consistency parameter is less than the second threshold are identified as the seven types of images.

16. The method according to claim 1, characterized in that, Obtain the set of historical images corresponding to the historical time period, including: Obtain the set of historical images corresponding to each of the consecutive historical time periods adjacent to the current time period.

17. The method according to claim 1, characterized in that, Update the target image set according to the first pixel correspondence and the second pixel correspondence of each target image, including: Determine the union of the first pixel correspondence and the second pixel correspondence of each target image to obtain the union of the relationships corresponding to each target image respectively; Generate the sparsity corresponding to the union of each relation to characterize the pixel defect rate; Determine the number of sparsities greater than a preset threshold. If the number is less than a preset number, resample the target image corresponding to the sparsity greater than the preset threshold to obtain a resampled image. The resampled image is added to the target image set.

18. The method according to claim 1, characterized in that, The current image processing model is trained based on the updated target image set and the historical image set to obtain the target image processing model, including: The current image processing model is trained based on the updated target image set, the historical image set, and the preset sample image set to obtain the target image processing model.

19. The method according to any one of claims 1 to 18, characterized in that, Also includes: Obtain the weights of the historical image processing model corresponding to the previous time period of the current time period; The weights of the current image processing model are updated based on the weights of the historical image processing models.

20. An evolution device for an image processing model, characterized in that, include: The image acquisition unit is used to acquire the set of target images corresponding to the current time period; The image acquisition unit is also used to acquire a set of historical images corresponding to the historical time period; The model evolution unit is used to train the current image processing model based on the target image set and the historical image set to obtain the target image processing model. This includes: obtaining the first pixel correspondence of each target image in the target image set according to the current image processing model corresponding to the current time period; obtaining the second pixel correspondence of each target image according to the original image processing model; updating the target image set according to the first pixel correspondence and the second pixel correspondence of each target image; and training the current image processing model based on the updated target image set and the historical image set to obtain the target image processing model.

21. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-19.

22. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to perform the method of any one of claims 1-19 by executing the executable instructions.

Images