Intelligent image processing method and system based on AI

Abstract

The invention provides an intelligent image processing method and system based on AI, and relates to the field of artificial intelligence, and the method comprises the steps: obtaining the image information of a to-be-processed image; obtaining a service target of the user; determining an image processing difficulty coefficient according to the image information; determining an image processing value coefficient according to the business target; determining a resource allocation parameter according to the image information, the image processing difficulty coefficient and the image processing value coefficient; and determining an image processing scheme according to the resource allocation parameter and the image processing value coefficient. According to the method and the device, the image processing scheme can be determined according to the resource allocation parameters and the image processing value coefficients during large-batch image processing, the processing efficiency of large-batch image processing can be effectively improved, and system computing power resources can be quickly released.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an AI-based intelligent image processing method and system. Background Art

[0002] In related technologies, images can be processed based on artificial intelligence models to improve the effect of image processing. However, related technologies do not consider the impact of image processing difficulty and image value on system processing resource allocation when processing large batches of images. That is, it is impossible to accurately generate image processing solutions based on image processing difficulty and image value.

[0003] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the Invention

[0004] The present invention provides an AI-based intelligent image processing method and system, which can solve the technical problem that related technologies are unable to accurately generate image processing solutions based on image processing difficulty and image value.

[0005] According to a first aspect of the present invention, there is provided an AI-based intelligent image processing method, comprising:

[0006] Obtaining image information of an image to be processed;

[0007] Obtain the user's business goals;

[0008] determining an image processing difficulty coefficient based on the image information;

[0009] Determining an image processing value coefficient based on the business objectives;

[0010] determining resource allocation parameters according to the image information, the image processing difficulty coefficient, and the image processing value coefficient;

[0011] An image processing solution is determined according to the resource allocation parameters and the image processing value coefficient.

[0012] According to the present invention, determining the image processing difficulty coefficient based on the image information includes:

[0013] Determining image resolution, noise intensity, total number of target pixels, and number of visible pixels based on the image information;

[0014] determining a peak signal-to-noise ratio according to the noise intensity;

[0015] Obtaining task difficulty recognition results for each image to be processed;

[0016] An image processing difficulty coefficient is determined according to the image resolution, the total number of target pixels, the number of visible pixels, the peak signal-to-noise ratio, and the task difficulty identification result.

[0017] According to the present invention, determining the image processing difficulty coefficient according to the image resolution, the total number of target pixels, the number of visible pixels, the peak signal-to-noise ratio, and the task difficulty identification result includes:

[0018] According to the formula

[0019]

[0020] Determine the image difficulty processing coefficient Dip of the i-th image to be processed i , where α1, α2, α3 and α4 are preset weights, Re i is the image resolution of the i-th image to be processed, Tnp i is the target total number of pixels of the i-th image to be processed, Vpc i is the number of visible pixels of the i-th image to be processed, Psn i is the peak signal-to-noise ratio of the i-th image to be processed, Tad i is the task difficulty recognition result of the i-th image to be processed, n is the number of images to be processed, i≤n, and both i and n are positive integers.

[0021] According to the present invention, determining the image processing value coefficient according to the business objective includes:

[0022] Acquiring annotation information according to the image information;

[0023] Determine the recognition result of the marked important pictures according to the marked information;

[0024] Obtain public user information;

[0025] Determine the user's important picture recognition results based on the public user information;

[0026] Determine the results of identifying important images in the content based on the business objectives;

[0027] An image processing value coefficient is determined according to the recognition results of the important labeled images, the recognition results of the important user images, and the recognition results of the important business images.

[0028] According to the present invention, determining the user's important picture recognition result based on the mass user information includes:

[0029] Determine the user click rate, user sharing rate, user collection rate and user download rate based on the mass user information;

[0030] Determining a user's high-frequency viewing recognition result based on the user's click rate and a preset user click rate threshold;

[0031] Determining a user high-frequency sharing identification result based on the user sharing rate and a preset user sharing rate threshold;

[0032] Determining a user's high-frequency collection recognition result based on the user's collection rate and a preset user collection rate threshold;

[0033] Determining a user high-frequency download identification result based on the user download rate and a preset user download rate threshold;

[0034] The user's important picture recognition results are determined based on the user's high-frequency viewing recognition results, the user's high-frequency sharing recognition results, the user's high-frequency collection recognition results, and the user's high-frequency download recognition results.

[0035] According to the present invention, determining the content-important picture recognition result according to the business goal includes:

[0036] The image to be processed is processed through the image description generation model to obtain the image semantic information;

[0037] Determine a business-related recognition result based on the image semantic information and the business goal.

[0038] According to the present invention, determining resource allocation parameters based on the image information, the image processing difficulty coefficient, and the image processing value coefficient includes:

[0039] determining an image size according to the image information;

[0040] determining an image collection time according to the image information;

[0041] Determining an image waiting time for processing according to the image collection time;

[0042] Resource allocation parameters are determined according to the image size, the image waiting processing time, the image processing difficulty coefficient and the image processing value coefficient.

[0043] According to the present invention, determining resource allocation parameters based on the image size, the image waiting time, the image processing difficulty coefficient, and the image processing value coefficient includes:

[0044] According to the formula

[0045]

[0046] Determine the resource allocation parameter Ra of the i-th image i , where β1 and β2 are preset weights, Vcp i is the image processing value coefficient of the i-th image to be processed, Wtp i is the image waiting time for processing the i-th image to be processed, Dip i is the image processing difficulty coefficient of the i-th image to be processed, Sp i is the image size of the i-th image to be processed, n is the number of images to be processed, i≤n, and both i and n are positive integers.

[0047] According to the present invention, determining an image processing solution based on the resource allocation parameter and the image processing value coefficient includes:

[0048] determining a processing model type according to the image processing value coefficient;

[0049] Determining a processing power allocation plan based on the resource allocation parameters;

[0050] An image processing solution is determined based on the type of processing model and the processing computing power allocation solution.

[0051] According to a second aspect of the present invention, there is provided an AI-based intelligent image processing system, comprising:

[0052] An image information module, used to obtain image information of an image to be processed;

[0053] Business information module, used to obtain the user's business goals;

[0054] a processing difficulty module, configured to determine an image processing difficulty coefficient based on the image information;

[0055] A processing value module, configured to determine an image processing value coefficient based on the business objective;

[0056] a resource allocation module, configured to determine resource allocation parameters according to the image information, the image processing difficulty coefficient, and the image processing value coefficient;

[0057] The processing solution module determines the image processing solution according to the resource allocation parameters and the image processing value coefficient.

[0058] Technical effect: According to the present invention, the image information of the image to be processed and the business goal of the user can be accurately obtained, and the image processing difficulty coefficient and image processing value coefficient of the image to be processed can be accurately analyzed based on the image information and business goals. Furthermore, the resource allocation parameters can be determined based on the image information, the image processing difficulty coefficient and the image processing value coefficient. When performing large-batch image processing, the image processing scheme can be determined based on the resource allocation parameters and the image processing value coefficient, effectively improving the processing efficiency of large-batch image processing and quickly releasing system computing resources. When determining the image processing difficulty coefficient, the image processing difficulty coefficient can be determined based on the image resolution, the total number of target pixels, the number of visible pixels, the peak signal-to-noise ratio and the task difficulty identification result. During the calculation process, the image difficulty processing of the image to be processed can be evaluated based on four aspects: resolution status, masking degree status, peak signal-to-noise ratio status and task difficulty status, thereby improving the comprehensiveness and accuracy of the image difficulty processing coefficient. When determining resource allocation parameters, the resource allocation parameters can be determined based on the image size, image waiting time, image processing difficulty coefficient and image processing value coefficient. During the calculation process, the resource allocation parameters are determined based on three aspects: the occupation of system resources by the image when it is not processed, the system resources called for processing and the relative importance. While giving priority to images that require less system resources for processing and images that are relatively more important, it can also avoid the long waiting time for processing of images that require more system resources for processing and images that are relatively less important, thereby improving the comprehensiveness and accuracy of the resource allocation parameters.

[0059] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and not limiting of the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can derive other embodiments based on these drawings without inventive efforts.

[0061] Figure 1 A schematic diagram exemplarily illustrates a flow chart of an AI-based intelligent image processing method according to an embodiment of the present invention;

[0062] Figure 2 A schematic diagram exemplarily illustrates a method for determining an image processing difficulty coefficient according to an embodiment of the present invention;

[0063] Figure 3A schematic diagram exemplarily illustrates a method for determining an image processing value coefficient according to an embodiment of the present invention;

[0064] Figure 4 A schematic diagram exemplarily illustrates a method for determining resource allocation parameters according to an embodiment of the present invention;

[0065] Figure 5 A schematic diagram exemplarily illustrates a method for determining an image processing solution according to an embodiment of the present invention;

[0066] Figure 6 A block diagram of an AI-based intelligent image processing system according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0068] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0069] Figure 1 A flowchart of an AI-based intelligent image processing method according to an embodiment of the present invention is exemplarily shown. The method includes:

[0070] Step S1, obtaining image information of an image to be processed;

[0071] Step S2, obtaining the user's business goals;

[0072] Step S3, determining an image processing difficulty coefficient based on the image information;

[0073] Step S4, determining an image processing value coefficient according to the business goal;

[0074] Step S5, determining resource allocation parameters according to the image information, the image processing difficulty coefficient, and the image processing value coefficient;

[0075] Step S6: Determine an image processing solution based on the resource allocation parameters and the image processing value coefficient.

[0076] According to the AI-based intelligent image processing method of an embodiment of the present invention, the image information of the image to be processed and the user's business goals can be accurately obtained, and the image processing difficulty coefficient and image processing value coefficient of the image to be processed can be accurately analyzed based on the image information and business goals. Furthermore, the resource allocation parameters can be determined based on the image information, the image processing difficulty coefficient and the image processing value coefficient. When performing large-batch image processing, the image processing plan can be determined based on the resource allocation parameters and the image processing value coefficient, thereby effectively improving the processing efficiency of large-batch image processing and quickly releasing system computing resources.

[0077] According to one embodiment of the present invention, in step S1, image information of an image to be processed is acquired.

[0078] For example, image information of the image to be processed, such as noise intensity, image size, and image resolution, can be obtained through professional tools (eg, NoiseLevelAnalysis, Pillow).

[0079] According to one embodiment of the present invention, in step S2, the user's business goal is obtained.

[0080] For example, determine the business goals of users who need to process images, such as the financial field or the sports field.

[0081] According to an embodiment of the present invention, in step S3, the image processing difficulty coefficient is determined based on the image information.

[0082] Figure 2 A schematic diagram of determining an image processing difficulty coefficient according to an embodiment of the present invention is exemplarily shown.

[0083] According to one embodiment of the present invention, step S3 includes:

[0084] Step S31, determining image resolution, noise intensity, target total pixel count, and visible pixel count based on the image information;

[0085] Step S32, determining a peak signal-to-noise ratio according to the noise intensity;

[0086] Step S33, obtaining the task difficulty recognition result of each image to be processed;

[0087] Step S34: determining an image processing difficulty coefficient based on the image resolution, the total number of target pixels, the number of visible pixels, the peak signal-to-noise ratio, and the task difficulty identification result.

[0088] For example, use image viewing software (such as Photoshop) to directly obtain image resolution, obtain image noise intensity based on the regional uniformity method, use a segmentation model (such as Mask R-CNN) to obtain the original mask of the image, that is, the total number of target pixels, and generate a mask for the same target in the occluded image, that is, the number of visible pixels; calculate the peak signal-to-noise ratio according to the calculation formula of the peak signal-to-noise ratio and the noise intensity. The lower the peak signal-to-noise ratio, the greater the noise and distortion, and the worse the image quality; obtain the task difficulty recognition result of each image to be processed, for example, when the processing task of the i-th image to be processed is to crop the image, apply basic filters, and compress the image at low resolution, it is a low difficulty task, then the task difficulty recognition result of the i-th image to be processed is 0; when the processing task of the i-th image to be processed is to perform high-resolution reconstruction, high-precision restoration, and complex scene segmentation on the image, it is a high difficulty task, then the task difficulty recognition result of the i-th image to be processed is 1; based on the image resolution, total number of target pixels, number of visible pixels, peak signal-to-noise ratio, and task difficulty recognition result, evaluate the task processing difficulty of the image and determine the image processing difficulty coefficient.

[0089] According to one embodiment of the present invention, step S34 includes: determining the image difficulty processing coefficient Dip of the i-th image to be processed according to formula (2): i ,

[0090]

[0091] Among them, α1, α2, α3 and α4 are preset weights, Re i is the image resolution of the i-th image to be processed, Tnp i is the target total number of pixels of the i-th image to be processed, Vpc i is the number of visible pixels of the i-th image to be processed, Psn i is the peak signal-to-noise ratio of the i-th image to be processed, Tad i is the task difficulty recognition result of the i-th image to be processed, n is the number of images to be processed, i≤n, and both i and n are positive integers.

[0092] According to one embodiment of the present invention, It is the ratio of the image resolution of the i-th image to be processed to the average image resolution of n images to be processed. The larger the ratio is, the higher the image resolution of the i-th image to be processed is. The higher the resolution is, the more high-frequency details are contained in the image, and the more difficult it is to process. is the ratio of the number of visible pixels of the i-th image to be processed to the target total number of pixels of the i-th image to be processed, Indicates the degree of masking of the i-th image to be processed, is the ratio of the masking degree of the i-th image to be processed to the average masking degree of the n-th image to be processed. The larger the ratio is, the higher the masking degree of the i-th image to be processed is. The higher the masking degree is, the more serious the information loss is and the more difficult it is to process. is the ratio of the average peak signal-to-noise ratio of n images to be processed to the peak signal-to-noise ratio of the i-th image to be processed. The larger the ratio, the smaller the peak signal-to-noise ratio of the i-th image to be processed. The smaller the peak signal-to-noise ratio, the worse the image quality, the more serious the distortion, and the greater the processing difficulty. i is the task difficulty recognition result of the i-th image to be processed. When the processing task of the i-th image to be processed is a high-difficulty task (such as high-resolution reconstruction, high-precision restoration and complex scene segmentation), Tad i The value of Tad is 1, which means that the processing difficulty is high. When the processing task of the i-th image to be processed is a low-difficulty task (such as cropping, basic filtering and low-resolution compression), Tad i A value of 0 indicates a low processing difficulty.

[0093] According to one embodiment of the present invention, To evaluate the image difficulty processing coefficient of the i-th image to be processed according to the resolution condition, the degree of masking condition, the peak signal-to-noise ratio condition and the task difficulty condition.

[0094] In this way, the image processing difficulty coefficient can be determined based on the image resolution, total number of target pixels, number of visible pixels, peak signal-to-noise ratio and task difficulty recognition results. During the calculation process, the image difficulty processing of the image to be processed can be evaluated according to four aspects: resolution condition, masking degree condition, peak signal-to-noise ratio condition and task difficulty condition, thereby improving the comprehensiveness and accuracy of the image difficulty processing coefficient.

[0095] According to one embodiment of the present invention, in step S4, an image processing value coefficient is determined according to the business goal.

[0096] Figure 3 A schematic diagram of determining an image processing value coefficient according to an embodiment of the present invention is exemplarily shown.

[0097] According to one embodiment of the present invention, step S4 includes:

[0098] Step S41, obtaining annotation information according to the image information;

[0099] Step S42: determining the recognition result of the marked important pictures according to the marked information;

[0100] Step S43, obtaining public user information;

[0101] Step S44, determining the user's important picture recognition result based on the public user information;

[0102] Step S45, determining the content important picture recognition result according to the business goal;

[0103] Step S46: determining an image processing value coefficient based on the recognition results of the important labeled images, the recognition results of the important user images, and the recognition results of the important business images.

[0104] For example, labeling information of the image to be processed is obtained through a labeling tool (such as Label Studio); if the image to be processed is marked as an important image (such as an urgent task and a paid service) in the labeling information of the image to be processed, the labeled important image recognition result of the image to be processed is 1, otherwise, the labeled important image recognition result of the image to be processed is 0; other users' access information to the image (such as click-through rate, sharing volume, and collection volume) is obtained from the collection source of the image to be processed (such as a public website), that is, the public user information; based on the public user information, the importance of the image is evaluated to determine the user important image recognition result; based on the business objectives, the importance of the image is evaluated to determine the content important image recognition result; based on the sum of the labeled important image recognition result, the user important image recognition result, and the business important image recognition result, the image processing value coefficient is determined.

[0105] According to one embodiment of the present invention, step S44 includes:

[0106] Step S441, determining the user click rate, user sharing rate, user collection rate and user download rate based on the public user information;

[0107] Step S442: determining the user's high-frequency viewing recognition results based on the user click rate and a preset user click rate threshold;

[0108] Step S443, determining a user high-frequency sharing identification result according to the user sharing rate and a preset user sharing rate threshold;

[0109] Step S444: determining a user's high-frequency collection recognition result based on the user's collection rate and a preset user collection rate threshold;

[0110] Step S445, determining a user high-frequency download identification result based on the user download rate and a preset user download rate threshold;

[0111] Step S446 , determining the user's important picture recognition results based on the user's high-frequency viewing recognition results, the user's high-frequency sharing recognition results, the user's high-frequency collection recognition results, and the user's high-frequency download recognition results.

[0112] For example, the user click rate, user sharing rate, user collection rate and user download rate are obtained through the dashboard building tools (such as Metabase, Tableau and Grafana); if the user click rate is greater than or equal to the preset user click rate threshold, it means that the click rate of the image is high, and the user high-frequency viewing recognition result is 1, otherwise, the user high-frequency viewing recognition result is 0, and the preset user click rate threshold can be set to 20%; if the user sharing rate is greater than or equal to the preset user sharing rate threshold, it means that the sharing rate of the image is high, and the user high-frequency sharing recognition result is 1, otherwise, the user high-frequency sharing recognition result is 0, and the preset user sharing rate threshold can be set to 2%; if the user collection rate is greater than or equal to the preset user collection rate threshold, it means that the collection rate of the image is high, and the user high-frequency collection recognition result is 1, otherwise, the user high-frequency collection recognition result is 0, and the preset user The collection rate threshold can be set to 5%; if the user download rate is greater than the preset user download rate threshold, it means that the download rate of the image is high, and the user high-frequency download recognition result is 1, otherwise, the user high-frequency download recognition result is 0. The preset user download rate threshold can be set to 15%; a high click-through rate indicates that the image has high visual appeal, a high sharing rate indicates that the image has use value, a high collection rate indicates that the image has high content depth, and a high download rate indicates that the image has high offline use needs. The user high-frequency viewing recognition results, user high-frequency sharing recognition results, user high-frequency collection recognition results and user high-frequency download recognition results are summed. If the sum value is greater than or equal to 2, it means that more users click, share, collect and download the image, and the image value of the image is high, then the user important picture recognition result is 1. If the sum value is less than 2, the user important picture recognition result is 0.

[0113] According to one embodiment of the present invention, step S45 includes:

[0114] Step S451, processing the image to be processed by the image description generation model to obtain image semantic information;

[0115] Step S452: Determine a business-related recognition result based on the image semantic information and the business goal.

[0116] For example, the image to be processed is processed through the image description generation model to obtain the image description information, that is, the image semantic information; the image semantic information and the business goal are converted into vector form. If the cosine similarity between the two vectors is greater than or equal to 0.5, it means that the image semantic information and the business goal are related, and the business-related recognition result is 10. Otherwise, the business-related recognition result is 0.

[0117] According to one embodiment of the present invention, in step S5, resource allocation parameters are determined based on the image information, the image processing difficulty coefficient, and the image processing value coefficient.

[0118] Figure 4 The following is a schematic diagram exemplarily showing a method for determining resource allocation parameters according to an embodiment of the present invention.

[0119] According to one embodiment of the present invention, step S5 includes:

[0120] Step S51, determining the image size according to the image information;

[0121] Step S52, determining the image collection time according to the image information;

[0122] Step S53, determining the image waiting time for processing according to the image collection time;

[0123] Step S54: determining resource allocation parameters according to the image size, the image waiting processing time, the image processing difficulty coefficient, and the image processing value coefficient.

[0124] For example, view the image size and image creation time, i.e., image collection time, through the image's operating system properties or command line tools; determine the image waiting time based on the current time minus the image collection time; evaluate the processing resource allocation status of the image to be processed based on the image size, image waiting time, image processing difficulty coefficient, and image processing value coefficient, and determine the resource allocation parameters.

[0125] According to one embodiment of the present invention, step S54 includes: determining the resource allocation parameter Ra of the i-th image according to formula (2): i ,

[0126]

[0127] Among them, β1 and β2 are preset weights, Vip i is the image processing value coefficient of the i-th image to be processed, Wtp i is the image waiting time for processing the i-th image to be processed, Dip i is the image processing difficulty coefficient of the i-th image to be processed, Sp i is the image size of the i-th image to be processed, n is the number of images to be processed, i≤n, and both i and n are positive integers.

[0128] According to one embodiment of the present invention, The ratio of the image waiting time of the i-th image to be processed to the average image waiting time of n images to be processed. The larger the ratio, the longer the image waiting time of the i-th image to be processed will be. This will cause the image to remain unprocessed for a long time, occupying storage space and wasting system resources. is the ratio of the image size of the i-th image to be processed to the average image size of n images to be processed. The larger the ratio is, the larger the image size of the i-th image to be processed is, and the larger the storage space occupied by the i-th image is. Indicates that based on the image waiting time and image size of the i-th image to be processed, the system resource usage of the i-th image to be processed is determined if it cannot be processed. The larger the value, the more serious the usage. It is the ratio of the image processing difficulty coefficient of the i-th image to be processed to the average image processing difficulty coefficient of n images to be processed. The larger the ratio is, the greater the image processing difficulty coefficient of the i-th image to be processed is, and the more system resources (such as computing power) are required to process the image. It is the ratio of the sum of the system resource occupation status of the i-th image to be processed and the system resources required for processing to the system resources required for processing, which represents the processing response ratio of the i-th image to be processed. The larger the ratio is, the more serious the system resource occupation status of the i-th image to be processed is, or the fewer system resources are required to process the image. The image is given priority processing. While giving priority to images requiring less system resources for processing, it can also avoid the long waiting time for images requiring more system resources to be processed, and can quickly release system computing resources. It is the relative difference between the image processing value coefficient of the i-th image to be processed and the average image processing value coefficient of n images to be processed, indicating the relative importance of the i-th image to be processed. The larger the ratio is, the higher the image processing value coefficient of the i-th image to be processed is, and the higher the importance of the image to be processed is. Giving priority to the image to be processed can enable the relatively more important images to be processed to be processed first, thereby meeting the needs of users.

[0129] In this way, resource allocation parameters can be determined based on image size, image waiting time, image processing difficulty coefficient and image processing value coefficient. During the calculation process, resource allocation parameters are determined based on three aspects: the occupation of system resources by the image when it is not processed, the system resources called for processing and the relative importance. While giving priority to images that require less system resources and images that are relatively more important, it can also avoid the long waiting time for images that require more system resources and images that are relatively less important, thereby improving the comprehensiveness and accuracy of resource allocation parameters.

[0130] According to one embodiment of the present invention, in step S6, an image processing solution is determined based on the resource allocation parameters and the image processing value coefficient.

[0131] Figure 5 A schematic diagram of determining an image processing solution according to an embodiment of the present invention is exemplarily shown.

[0132] According to one embodiment of the present invention, step S6 includes:

[0133] Step S61, determining the type of processing model according to the image processing value coefficient;

[0134] Step S62: determining a processing power allocation plan based on the resource allocation parameters;

[0135] Step S63: Determine an image processing solution based on the processing model type and the processing computing power allocation solution.

[0136] For example, if the image processing value coefficient of an image is greater than or equal to 10, it is a high-importance image and is processed using a large model (such as ResNet-152); if the image processing value coefficient of an image is less than 10, it is a low-importance image and is processed using a small model (such as MobileNet); the resource allocation parameters of all images to be processed are arranged in descending order to generate a resource allocation parameter sequence, and more GPU cores are allocated to the images to be processed corresponding to the first half of the elements in the resource allocation parameter sequence, such as reserving at least 20% of streaming multiprocessors for the images to be processed, and fewer GPU cores are allocated to the images to be processed corresponding to the last half of the elements in the resource allocation parameter sequence, such as reserving at most 5% of streaming multiprocessors for the images to be processed; the image processing plan is determined based on the type of processing model (such as large model and small model) and the processing computing power allocation (such as GPU cores and processing power batches).

[0137] According to the AI-based intelligent image processing method of an embodiment of the present invention, the image information of the image to be processed and the business goal of the user can be accurately obtained, and the image processing difficulty coefficient and image processing value coefficient of the image to be processed can be accurately analyzed based on the image information and the business goal. Furthermore, the resource allocation parameters can be determined based on the image information, the image processing difficulty coefficient and the image processing value coefficient. When performing large-batch image processing, the image processing scheme can be determined based on the resource allocation parameters and the image processing value coefficient, effectively improving the processing efficiency of large-batch image processing and quickly releasing system computing resources. When determining the image processing difficulty coefficient, the image processing difficulty coefficient can be determined based on the image resolution, the total number of target pixels, the number of visible pixels, the peak signal-to-noise ratio and the task difficulty identification result. During the calculation process, the image difficulty processing of the image to be processed can be evaluated based on four aspects: resolution status, masking degree status, peak signal-to-noise ratio status and task difficulty status, thereby improving the comprehensiveness and accuracy of the image difficulty processing coefficient. When determining resource allocation parameters, the resource allocation parameters can be determined based on the image size, image waiting time, image processing difficulty coefficient and image processing value coefficient. During the calculation process, the resource allocation parameters are determined based on three aspects: the occupation of system resources by the image when it is not processed, the system resources called for processing and the relative importance. While giving priority to images that require less system resources for processing and images that are relatively more important, it can also avoid the long waiting time for processing of images that require more system resources for processing and images that are relatively less important, thereby improving the comprehensiveness and accuracy of the resource allocation parameters.

[0138] Figure 2 A block diagram of an AI-based intelligent image processing system according to an embodiment of the present invention is exemplarily shown, wherein the system includes:

[0139] An image information module, used to obtain image information of an image to be processed;

[0140] Business information module, used to obtain the user's business goals;

[0141] a processing difficulty module, configured to determine an image processing difficulty coefficient based on the image information;

[0142] A processing value module, configured to determine an image processing value coefficient based on the business objective;

[0143] a resource allocation module, configured to determine resource allocation parameters according to the image information, the image processing difficulty coefficient, and the image processing value coefficient;

[0144] The processing solution module determines the image processing solution according to the resource allocation parameters and the image processing value coefficient.

[0145] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0146] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from the principles described.

Claims

1. An AI-based intelligent image processing method, characterized in that: include: Obtaining image information of an image to be processed; Obtain the user's business goals; determining an image processing difficulty coefficient based on the image information; Determining an image processing value coefficient based on the business objectives; determining resource allocation parameters according to the image information, the image processing difficulty coefficient, and the image processing value coefficient; An image processing solution is determined according to the resource allocation parameters and the image processing value coefficient.

2. The AI-based intelligent image processing method according to claim 1, characterized in that: Determining an image processing difficulty coefficient according to the image information includes: Determining image resolution, noise intensity, total number of target pixels, and number of visible pixels based on the image information; determining a peak signal-to-noise ratio according to the noise intensity; Obtaining task difficulty recognition results for each image to be processed; An image processing difficulty coefficient is determined according to the image resolution, the total number of target pixels, the number of visible pixels, the peak signal-to-noise ratio, and the task difficulty identification result.

3. The AI-based intelligent image processing method according to claim 2, characterized in that: Determining an image processing difficulty coefficient according to the image resolution, the total number of target pixels, the number of visible pixels, the peak signal-to-noise ratio, and the task difficulty identification result, including: According to the formula Determine the image difficulty processing coefficient Dip of the i-th image to be processed i , where α1, α2, α3 and α4 are preset weights, Re i is the image resolution of the i-th image to be processed, Tnp i is the target total number of pixels of the i-th image to be processed, Vpc i is the number of visible pixels of the i-th image to be processed, Psn i is the peak signal-to-noise ratio of the i-th image to be processed, Tad i is the task difficulty recognition result of the i-th image to be processed, n is the number of images to be processed, i≤n, and both i and n are positive integers.

4. The AI-based intelligent image processing method according to claim 1, characterized in that: Based on the business objectives, determine the image processing value factor, including: Acquiring annotation information according to the image information; Determine the recognition result of the marked important pictures according to the marked information; Obtain public user information; Determine the user's important picture recognition results based on the public user information; Determine the results of identifying important images in the content based on the business objectives; An image processing value coefficient is determined according to the recognition results of the important labeled images, the recognition results of the important user images, and the recognition results of the important business images.

5. The AI-based intelligent image processing method according to claim 4, characterized in that: Determine the user's important image recognition results based on the public user information, including: Determine the user click rate, user sharing rate, user collection rate and user download rate based on the mass user information; Determining a user's high-frequency viewing recognition result based on the user's click rate and a preset user click rate threshold; Determining a user high-frequency sharing identification result based on the user sharing rate and a preset user sharing rate threshold; Determining a user's high-frequency collection recognition result based on the user's collection rate and a preset user collection rate threshold; Determining a user high-frequency download identification result based on the user download rate and a preset user download rate threshold; The user's important picture recognition results are determined based on the user's high-frequency viewing recognition results, the user's high-frequency sharing recognition results, the user's high-frequency collection recognition results, and the user's high-frequency download recognition results.

6. The AI-based intelligent image processing method according to claim 4, characterized in that: Based on the business objectives, determine the results of identifying important images in the content, including: The image to be processed is processed through the image description generation model to obtain the image semantic information; Determine a business-related recognition result based on the image semantic information and the business goal.

7. The AI-based intelligent image processing method according to claim 1, characterized in that: Determining resource allocation parameters according to the image information, the image processing difficulty coefficient, and the image processing value coefficient includes: determining an image size according to the image information; determining an image collection time according to the image information; Determining an image waiting time for processing according to the image collection time; Resource allocation parameters are determined according to the image size, the image waiting processing time, the image processing difficulty coefficient and the image processing value coefficient.

8. The AI-based intelligent image processing method according to claim 7, characterized in that: Determining resource allocation parameters according to the image size, the image waiting time, the image processing difficulty coefficient, and the image processing value coefficient includes: According to the formula Determine the resource allocation parameter Ra of the i-th image i , where β1 and β2 are preset weights, Vip i is the image processing value coefficient of the i-th image to be processed, Wtp i is the image waiting time for processing the i-th image to be processed, Dip i is the image processing difficulty coefficient of the i-th image to be processed, Sp i is the image size of the i-th image to be processed, n is the number of images to be processed, i≤n, and both i and n are positive integers.

9. The AI-based intelligent image processing method according to claim 1, characterized in that: Determining an image processing solution according to the resource allocation parameter and the image processing value coefficient includes: determining a processing model type according to the image processing value coefficient; Determining a processing power allocation plan based on the resource allocation parameters; An image processing solution is determined based on the type of processing model and the processing computing power allocation solution.

10. An AI-based intelligent image processing system, characterized in that: include: An image information module, used to obtain image information of an image to be processed; Business information module, used to obtain the user's business goals; a processing difficulty module, configured to determine an image processing difficulty coefficient based on the image information; A processing value module, configured to determine an image processing value coefficient based on the business objective; a resource allocation module, configured to determine resource allocation parameters according to the image information, the image processing difficulty coefficient, and the image processing value coefficient; The processing solution module determines the image processing solution according to the resource allocation parameters and the image processing value coefficient.

Images