Image feature library updating method, image feature library checking method, and related device

Abstract

This application discloses an image feature library update method, a verification method, and related equipment. The image feature library update method includes: executing an image processing task at least once using an image processing model until the image feature pool meets preset requirements; the image processing task includes: performing target image processing on the image to be processed using the image processing model to obtain image information of the image to be processed, and adding the image information to the image feature pool, the image information including at least the image features of the image to be processed; updating the basic image feature library associated with the target image processing based on the image feature pool that meets the preset requirements, wherein the updated basic image feature library is used to verify the image processing result obtained by the image processing model performing target image processing on the image. Through the above method, this application can adaptively update the basic image feature library, improving the applicability of the basic image feature library used for verification.

Description

Technical Field

[0001] This application relates to the field of image processing technology, specifically to an image feature library update method, verification method, and related equipment. Background Technology

[0002] Currently, image processing models may detect false positives. Identifying these false positives primarily involves comparing the image features in the original image with features in an offline-built image feature database. However, the current image feature database is built offline, requiring manual updates based on different application scenarios, making adaptive updates impossible. Summary of the Invention

[0003] This application provides a method for updating an image feature library, a method for verifying an image feature library, and related equipment, which can adaptively update the basic image feature library and improve the applicability of the basic image feature library used for verification.

[0004] To address the aforementioned technical problems, the technical solution adopted in this application is as follows: A method for updating an image feature library for verification is provided. This method includes: executing an image processing task at least once using an image processing model until the image feature pool meets preset requirements; the image processing task includes: performing target image processing on the image to be processed using the image processing model to obtain image information of the image to be processed, and adding the image information to the image feature pool, wherein the image information includes at least the image features of the image to be processed; and updating the basic image feature library associated with the target image processing based on the image feature pool that meets the preset requirements, wherein the updated basic image feature library is used to verify the image processing results obtained by the image processing model performing target image processing on the image.

[0005] To solve the above-mentioned technical problems, another technical solution adopted in this application is: to provide an image feature library update device, the device including a memory and a processor connected to each other, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, it is used to implement the image feature library update method for verification in the above-mentioned technical solution.

[0006] To address the aforementioned technical problems, another technical solution adopted in this application is: providing a method for verifying image processing results. This method includes: performing target image processing on the image to be processed using an image processing model to obtain the image processing result; acquiring a basic image feature library associated with the target image processing, wherein the basic image feature library is obtained through the image feature library update method for verification in the aforementioned technical solution; and verifying the image processing result based on the basic image feature library.

[0007] To solve the above-mentioned technical problems, another technical solution adopted in this application is: to provide an image processing result verification device, the device including a memory and a processor connected to each other, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, it is used to implement the image processing result verification method in the above-mentioned technical solution.

[0008] To solve the above-mentioned technical problems, another technical solution adopted in this application is to provide a computer-readable storage medium for storing a computer program. When the computer program is executed by a processor, it is used to implement the method for updating the image feature library for verification or the method for verifying image processing results in the above-mentioned technical solution.

[0009] The beneficial effects of this application through the above scheme are as follows: by executing an image processing task at least once through an image processing model until the image feature pool meets the preset requirements, and then updating the basic image feature library associated with the target image processing based on the image feature pool that meets the preset requirements, this embodiment can adaptively update the basic image feature library according to the real-time image processing results of the image processing model set under different application scenarios, so that the basic image feature library can automatically adapt to various application scenarios according to actual needs, greatly improving the applicability and flexibility of the basic image feature library used for verification, and eliminating the need for manual maintenance of the basic image feature library, greatly reducing manual maintenance costs and improving the update efficiency of the basic image feature library used for verification. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0011] Figure 1 This is a flowchart illustrating an embodiment of the image feature library update method for verification provided in this application;

[0012] Figure 2 This is a flowchart illustrating another embodiment of the image feature library update method for verification provided in this application;

[0013] Figure 3 This is a flowchart illustrating an embodiment of step 22 provided in this application;

[0014] Figure 4 This is a schematic diagram of an embodiment of the image feature library updating device provided in this application;

[0015] Figure 5 This is a flowchart illustrating an embodiment of the image processing result verification method provided in this application;

[0016] Figure 6 This is a schematic diagram of an embodiment of the image processing result verification device provided in this application;

[0017] Figure 7 This is a schematic diagram of an embodiment of the computer-readable storage medium provided in this application. Detailed Implementation

[0018] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the application. Similarly, the following embodiments are only some, not all, embodiments of the present application, and all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present application.

[0019] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] It should be noted that the terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0021] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the image feature library update method for verification provided in this application. The method includes:

[0022] Step 11: Perform an image processing task at least once using the image processing model until the image feature pool meets the preset requirements.

[0023] The image processing model performs at least one image processing task until the image feature pool meets the preset requirements. The image processing task includes: performing target image processing on the image to be processed through the image processing model to obtain image information of the image to be processed, and adding the image information to the image feature pool. The image information includes at least the image features of the image to be processed.

[0024] Specifically, the image processing model can be a trained deep learning network model with image processing capabilities, such as a wearable detection model, a worker detection model, or a vehicle attribute detection model. Different image processing models can be obtained or trained according to the actual application, and the type of image processing model is not limited here. Taking a face detection model as an example, the image information obtained by the face detection model through target image processing of the image to be processed may include facial features in the image to be processed.

[0025] Step 12: Update the basic image feature library associated with the target image processing based on the image feature pool that meets the preset requirements.

[0026] Based on an image feature pool that meets preset requirements, the basic image feature library associated with the target image processing is updated. The updated basic image feature library is used to verify the image processing results obtained by the image processing model when processing the target image, to determine whether the image processing model has misjudged the target image. Specifically, the image features stored in the updated basic image feature library can be compared with the image features generated by the image processing model to calculate their similarity. A similarity value is then used to determine whether the image processing model has misjudged the target image. For example, a similarity value less than a preset threshold indicates that the image processing model has misjudged the target image.

[0027] In one embodiment, the image processing model can perform multi-task target image processing on the image to be detected, that is, the target image processing can include multiple sub-processing tasks. In this case, the basic image feature library associated with the target image processing can include multiple sub-basic image feature libraries. The number of sub-basic image feature libraries is the same as the number of sub-processing tasks, and the sub-basic image feature libraries associated with different sub-processing tasks are different. It can be understood that when the image processing model performs single-task target image processing on the image to be detected, the number of basic image feature libraries associated with the target image processing is one.

[0028] Taking the image processing model as an example of a wear detection model that performs wear detection tasks, it can be used to detect the upper garment, lower garment, and headwear of the human body in the image to be processed. In this case, the wear detection task performed by the wear detection model includes three sub-detection tasks: upper garment detection task, lower garment detection task, and headwear detection task. At this time, the basic image feature library can be divided into three sub-basic image feature libraries that correspond one-to-one with the sub-detection tasks. The sub-basic image feature libraries can then be used to verify the image information generated by the wear detection model in performing the corresponding sub-detection tasks.

[0029] Taking the application of a basic image feature library in the security field as an example, the image processing model can be a suspicious person detection model used to detect suspicious persons in the image to be processed. When the image processing model detects a suspicious person, it can trigger an alarm function to alert security personnel. The basic image feature library obtained by the image feature library update method proposed in this embodiment can be adaptively updated based on the feature information of suspicious persons obtained by the suspicious person detection model. Then, the updated basic image feature library is used to verify the feature information of suspicious persons that are about to trigger an alarm, thereby determining whether the image processing model has made a false judgment. If it is determined that the image processing model has made a false judgment, the alarm is canceled; if it is determined that the image processing model has not made a false judgment, the alarm continues. Through the above methods, the false alarm rate of the suspicious person detection model can be greatly reduced, and the accuracy of security can be improved. It is understood that the basic image feature library obtained by the image feature library update method for verification proposed in this embodiment can be widely applied to various scenarios according to actual needs, and examples are not given here.

[0030] This embodiment executes an image processing task at least once through an image processing model until the image feature pool meets preset requirements. Then, based on the image feature pool that meets the preset requirements, the basic image feature library associated with the target image processing is updated. This embodiment can adaptively update the basic image feature library according to the real-time image processing results of the image processing model set under different application scenarios, so that the basic image feature library can automatically adapt to various application scenarios according to actual needs, greatly improving the applicability and flexibility of the basic image feature library used for verification. Moreover, no manual maintenance of the basic image feature library is required, greatly reducing manual maintenance costs and improving the update efficiency of the basic image feature library used for verification.

[0031] Please see Figure 2 , Figure 2 This is a flowchart illustrating another embodiment of the image feature library update method for verification provided in this application. The method includes:

[0032] Step 21: Perform an image processing task at least once using the image processing model until the image feature pool meets the preset requirements.

[0033] Step 21 is the same as step 11 in the above embodiments, and will not be repeated here. Specifically, in response to the number of image features stored in the image feature pool reaching a preset value, it is determined that the image feature pool meets the preset requirements. The preset value can be set according to the actual situation, and is not limited here.

[0034] In one embodiment, when the image processing model performs multi-task target image processing on the image to be detected, the image information generated by the image processing model may include the corresponding image features obtained by the image processing model from each sub-processing task performed on the image to be processed; the image feature pool is used to store image features, and the image feature pool may correspond one-to-one with the sub-processing task. The image feature pool can be used to store the corresponding image features generated by the image processing model when performing the sub-processing task.

[0035] Taking the wear detection model as an example, the image information obtained by the wear detection model from the image to be processed can include the upper garment feature information, lower garment feature information and headdress feature information of the human body in the image to be processed. Then, the upper garment feature information, lower garment feature information and headdress feature information can be stored in three image feature pools respectively.

[0036] Understandably, in the case of multi-task target image processing of the image to be detected by the above image processing model, it can be determined whether each image feature pool meets the preset requirements. That is, it can be determined whether the number of image features stored in each image feature pool reaches the preset value. Then, the image features stored in the image feature pool that reaches the preset value are used to update the corresponding sub-basic image feature library. The sub-basic image feature library corresponding to the image feature pool that does not reach the preset value is not updated.

[0037] Step 22: Based on the confidence level of the processing results contained in each image information in the image feature pool, determine the target image features from the image features contained in each image information in the image feature pool.

[0038] Image information also includes processing result confidence, which is the confidence level of the image processing result obtained by the image processing model from the target image. Based on the processing result confidence levels contained in each image information in the image feature pool, target image features can be determined from the image features contained in each image information in the image feature pool, thus selecting high-quality target image features based on confidence levels.

[0039] Step 23: Update the base image feature library associated with target image processing based on target image features.

[0040] The base image feature library associated with target image processing is updated based on the target image features. Specifically, the base image feature library associated with target image processing can be updated based on the target image features in the image feature pool; then, in response to the completion of the base image feature library update, the corresponding image features stored in the image feature pool are deleted.

[0041] In other words, when the number of image features stored in the image feature pool reaches a preset value, the basic image feature library associated with the target image processing is updated. After the basic image feature library is updated, the image features stored in the corresponding image feature pool are deleted, and then image features are re-stored in the image feature pool until the number of image features stored in the image feature pool reaches the preset value again. Then the basic image feature library is updated again, and so on.

[0042] This embodiment determines target image features from the image features contained in each image information in the image feature pool based on the confidence level of the processing results. It can use the confidence level to perform quality screening of image features, selecting high-quality target image features before updating the basic image feature library, greatly improving the quality of the basic image feature library used for verification, and thus improving the accuracy of subsequent verification. Furthermore, by setting corresponding image feature pools according to different sub-processing tasks, the collected image features are stored in the image feature pool of the corresponding sub-processing task. Whenever the number of image features stored in the image feature pool reaches a preset value, the corresponding sub-basic image feature library is updated. After each update of the basic image feature library, the corresponding image features in the image feature pool are deleted, greatly improving the update efficiency and flexibility of the basic image feature library.

[0043] Please see Figure 3 , Figure 3 This is a flowchart illustrating an embodiment of step 22. The method includes:

[0044] Step 31: Aggregate image features with corresponding scale information to obtain image feature vectors.

[0045] Image information also includes scale information corresponding to image features. By focusing on image features and their corresponding scale information, an image feature vector can be obtained. Here, scale information refers to the area proportion of the target object in the image to be processed. A larger area proportion indicates a large scale, and a smaller area proportion indicates a small scale. By retaining scale information and aggregating image features and their corresponding scale information, the feature differences between target objects of different scales can be increased, thereby improving the completeness and accuracy of the obtained image feature vector.

[0046] Specifically, methods for aggregating image features with their corresponding scale information may include, but are not limited to, feature mapping. Feature mapping is equivalent to dimensionality reduction, which may include principal component analysis (PCA) or multi-dimensional scaling. The main idea of ​​PCA is to map n-dimensional features onto k dimensions, where k dimensions are entirely new orthogonal features (i.e., principal components). In other words, high-dimensional features are projected onto principal component features, and then features with smaller variance are discarded, leaving the linear feature set with the largest variance. Multi-dimensional scaling uses the correlation between feature information for dimensionality reduction. Using multi-dimensional scaling preserves the main low-dimensional space of image features. In this low-dimensional space, image features maintain consistency with high-dimensional data information and preserve the difference in Euclidean distance between feature information. In other implementations, other conventional feature mapping methods in the technical field can be selected to complete the aggregation operation, and are not limited here.

[0047] Understandably, in one embodiment, the image feature vector generated by the image processing model when performing target image processing on the image to be processed can be directly acquired. In this case, step 31 can be omitted, and the process can proceed directly to step 32.

[0048] Step 32: Use a clustering algorithm to group the image feature vectors to obtain at least two feature groups.

[0049] Clustering algorithms can be used to group image feature vectors, resulting in at least two feature groups. Each feature group contains multiple image feature vectors and their corresponding confidence scores. Specifically, unsupervised clustering algorithms can be used to group image feature vectors. These algorithms can include, but are not limited to, K-means clustering, density-based spatial clustering (DBSCAN), or Gaussian mixture model clustering (GMM). Taking K-means clustering as an example, we first define all image feature vectors in the image feature pool as a set B = {b1, b2, ..., bn}. Then, we determine K samples as cluster centers, for example, selecting 3 cluster centers. We then calculate the distance between each sample (i.e., the image feature vector) and the cluster center, regressing each sample to the nearest cluster center. We calculate the mean of the samples corresponding to each cluster center, and find the sample corresponding to the mean as the new cluster center. This process is repeated until the clustering terminates when the cluster centers no longer change or the number of clustering iterations reaches the required number.

[0050] Step 33: Based on the confidence level in the feature group, select candidate feature groups from the feature group.

[0051] Based on the confidence level in the feature group, candidate feature groups are selected from the feature group. Specifically, in one embodiment, the mean and standard deviation of the confidence level in each feature group can be calculated, and at least the feature group with the largest mean or the smallest standard deviation can be selected as a candidate feature group. That is, one, two or more feature groups with the largest mean can be selected as candidate feature groups, or one, two or more feature groups with the smallest standard deviation can be selected as candidate feature groups. The number of candidate feature groups is not limited here.

[0052] A large mean indicates that the confidence level in the feature group is generally high, and the image feature vectors contained in the feature group are relatively accurate positive sample feature vectors; a small standard deviation indicates that the confidence level in the feature group fluctuates less, and the image feature vectors in the feature group are relatively stable; feature groups with large means or small standard deviations can be selected as candidate feature groups according to actual needs, in order to prepare for further screening of target image features.

[0053] In another embodiment, the mean and standard deviation of the confidence scores corresponding to the image feature vectors in each feature group can be calculated. Then, the mean and standard deviation are summed with weights. If the weight of the mean is greater than a preset ratio, at least the feature group with the largest sum is selected as a candidate feature group; if the weight of the standard deviation is greater than a preset ratio, at least the feature group with the smallest sum is selected as a candidate feature group. That is, when the weight of the mean is relatively large, one, two, or more feature groups with the largest sum can be selected as candidate feature groups; when the weight of the standard deviation is relatively large, one, two, or more feature groups with the smallest sum can be selected as candidate feature groups. The weight values ​​of the mean and standard deviation, as well as the preset ratio, can be set according to actual needs. In other embodiments, the weight settings can also be used to filter out positive sample feature vectors with smaller fluctuations and higher accuracy; this is not limited here.

[0054] Step 34: Based on the confidence level in the candidate feature group, select multiple target image features.

[0055] After selecting candidate feature groups from the feature groups, the image feature vectors in the candidate feature groups can be further filtered. Specifically, the confidence levels in the candidate feature groups are sorted according to their confidence scores. The image feature vectors whose confidence scores are ranked at a specified sequence position are selected as the target image features. The specified sequence position can be the first k1 positions sorted from largest to smallest confidence, or the last k2 positions sorted from smallest to largest confidence; or it can be the k3 positions starting from the median confidence level and moving towards the direction of increasing confidence when sorting from largest to smallest or smallest to largest confidence. The specific values ​​of the specified sequence position and k1-k3 can be set according to the actual situation and are not limited here.

[0056] Understandably, when there are two or more candidate feature groups, one, two or more image feature vectors at a specified sequence position can be selected from each candidate feature group as the target image feature.

[0057] The following section provides a detailed description of the content of updating the basic image feature library associated with the target image processing based on the target image features in step 23:

[0058] In one embodiment, step 23 may include: determining whether there are anomalous features in the target image features; in response to the presence of anomalous features in the target image features, updating at least a portion of the remaining target image features (excluding the anomalous features) to the base image feature library associated with the target image processing; and in response to the absence of anomalous features in the target image features, updating all target image features to the base image feature library associated with the target image processing. The anomalous features may be deleted, and then at least a portion of the remaining target image features may be updated to the base image feature library associated with the target image processing; alternatively, at least a portion of the desired target image features may be directly selected from the remaining target image features (excluding the anomalous features) and updated to the base image feature library associated with the target image processing. The actual execution method is not limited here.

[0059] Specifically, the steps to determine whether there are anomalous features in the target image features may include: calculating the cross-similarity between each target image feature in the same candidate feature group to obtain the similarity result; and then, based on the similarity result, using outlier analysis to determine whether there are anomalous features in the target image features.

[0060] The similarity between target image features can be calculated by methods such as Euclidean distance or cosine distance. Taking the cosine distance between target image features as an example, the similarity result can be obtained using the following formula:

[0061]

[0062] In the above formula, d cosθ Let X = {x1, x2, ..., xk} be the set of target image features in the same candidate feature group, and let D = {d11, d12, ..., dkk} be the similarity value between target image feature a and target image feature b.

[0063] The following is a brief explanation using the Isolation Forest method, an outlier analysis technique, to determine whether there are anomalous features in the target image features of each candidate feature group. Isolation Forest isolates each anomalous sample (i.e., anomalous feature) by constructing a binary tree. Specifically, it segments the similarity result D = {d11, d12, ..., dkk}. Since outliers are far from normal sample clusters, data that can be segmented with very few iterations can be considered anomalous samples. After the outlier analysis algorithm described above, it can be determined whether there are anomalous features in the target image features of each candidate feature group. If there are no anomalous features in the target image features of each candidate feature group, all target image features can be registered in the basic image feature library associated with target image processing. If there are anomalous features in the target image features of a candidate feature group, at least a portion of the remaining target image features (excluding the anomalous features) can be updated in the basic image feature library associated with target image processing. It is understood that in other embodiments, conventional outlier analysis algorithms in the field of outlier detection technology can be used to implement outlier analysis, and this is not limited here.

[0064] In one embodiment, the method for updating target image features (here referring to at least a portion of all target image features excluding anomalous features) to a base image feature library associated with target image processing may include: acquiring historical image features already stored in the base image feature library; caching historical image features and clearing the base image feature library; then sorting the confidence scores corresponding to the historical image features and the target image features by size, and selecting a preset number of feature vectors with the highest confidence scores to store in the base image feature library. The preset number can be set according to actual conditions and is not limited here.

[0065] In another embodiment, the method for updating target image features to a basic image feature library may further include: in response to the remaining storage space in the basic image feature library being less than the amount of data of the target image features to be stored, deleting some or all of the historical image features already stored in the basic image feature library, and storing the target image features to be stored into the basic image feature library; in response to the remaining storage space in the basic image feature library being greater than or equal to the amount of data of the target image features to be stored, storing the target image features to be stored into the basic image feature library.

[0066] This embodiment obtains image feature vectors by aggregating image features and scale information, which increases the feature differences between target objects at different scales and improves the completeness and accuracy of the obtained image feature vectors. Subsequent calculations using these image feature vectors significantly reduce computational load and improve efficiency, thereby increasing update efficiency. Furthermore, by repeatedly filtering the image feature vectors generated by the acquired image processing model, high-confidence and stable positive sample target image features are selected and updated to the basic image feature library. This enables adaptive updates to the basic image feature library while significantly improving the feature quality within the library, thus enhancing the accuracy of subsequent image information verification. Moreover, during the update of the image feature vectors to the basic image feature library, the library can be updated selectively based on existing historical image features and the target image features to be added, further ensuring the iterative quality of the basic image feature library.

[0067] Please see Figure 4 , Figure 4 This is a schematic diagram of an embodiment of the image feature library updating device provided in this application. The image feature library updating device 40 includes a memory 41 and a processor 42 connected to each other. The memory 41 is used to store a computer program. When the computer program is executed by the processor 42, it is used to implement the image feature library updating method for verification in the above embodiment.

[0068] Please see Figure 5 , Figure 5 This is a flowchart illustrating an embodiment of the image processing result verification method provided in this application. The method includes:

[0069] Step 51: Use an image processing model to perform target image processing on the image to be processed to obtain the image processing result.

[0070] The image processing model performs target image processing on the image to be processed to obtain the image processing result, which may contain the image features of the image to be processed.

[0071] Step 52: Obtain the basic image feature library associated with the target image processing.

[0072] The basic image feature library is obtained through the image feature library update method for verification described in the above embodiments.

[0073] Step 53: Verify the image processing results based on the basic image feature library.

[0074] The image processing results are verified based on the basic image feature library. Specifically, the target image features stored in the updated basic image feature library can be compared with the image features generated by the image processing model to calculate the similarity between the two and obtain the similarity value. Then, the similarity value is used to determine whether the image processing model has misjudged. For example, if the similarity value is less than a preset threshold, it can be indicated that the image processing model has misjudged. No limit is placed on the verification method here.

[0075] This embodiment utilizes the adaptively updated basic image feature library to perform false detection judgment on the image information subsequently generated by the image processing model, which can improve the verification efficiency and accuracy and reduce the false alarm rate of the image processing model.

[0076] Please see Figure 6 , Figure 6 This is a schematic diagram of an embodiment of the image processing result verification device provided in this application. The image processing result verification device 60 includes a memory 61 and a processor 62 connected to each other. The memory 61 is used to store a computer program. When the computer program is executed by the processor 62, it is used to implement the image processing result verification method in the above embodiment.

[0077] Please see Figure 7 , Figure 7 This is a schematic diagram of an embodiment of a computer-readable storage medium provided in this application. The computer-readable storage medium 70 is used to store a computer program 71. When the computer program 71 is executed by a processor, it is used to implement the image feature library update method or image processing result verification method in the above embodiment.

[0078] The computer-readable storage medium 70 can be any medium capable of storing program code, such as a server, USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0079] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0080] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0081] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0082] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

[0083] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for updating an image feature library for verification, characterized in that, include: The step of performing at least one image processing task through an image processing model until the image feature pool meets a preset requirement includes: in response to the number of image features stored in the image feature pool reaching a preset value, determining that the image feature pool meets the preset requirement; the image processing task includes: performing target image processing on the image to be processed through the image processing model to obtain image information of the image to be processed, and adding the image information to the image feature pool, wherein the image information includes at least the image features of the image to be processed and the confidence level of the processing result, and the confidence level of the processing result is the confidence level of the image processing result of the target image processing on the image to be processed by the image processing model; The step of updating the basic image feature library associated with the target image processing based on the image feature pool that meets the preset requirements includes: determining target image features from the image features contained in each image information in the image feature pool according to the confidence level of the processing result contained in each image information in the image feature pool; updating the basic image feature library associated with the target image processing based on the target image features; wherein, the updated basic image feature library is used to verify the image processing result obtained by the image processing model performing the target image processing on the image; The image information further includes scale information corresponding to the image features, whereby the scale information represents the area proportion of the target object in the image to be processed. The step of determining the target image features from the image features contained in each image information in the image feature pool based on the processing result confidence level contained in each image information in the image feature pool includes: aggregating the image features and the corresponding scale information to obtain an image feature vector; grouping the image feature vectors using a clustering algorithm to obtain at least two feature groups, wherein each feature group contains multiple image feature vectors and confidence levels; selecting candidate feature groups from the feature groups based on the confidence levels in the feature groups; and selecting multiple target image features based on the confidence levels in the candidate feature groups.

2. The method for updating the image feature library for verification according to claim 1, characterized in that, The step of updating the base image feature library associated with the target image processing based on the target image features includes: Based on the target image features in the image feature pool, update the basic image feature library associated with the target image processing; In response to the completion of the update of the basic image feature library, the corresponding image features stored in the image feature pool are deleted.

3. The method for updating the image feature library for verification according to claim 1, characterized in that, The step of updating the base image feature library associated with the target image processing based on the target image features includes: Determine whether there are any abnormal features in the target image features; In response to the presence of abnormal features in the target image features, at least a portion of the remaining target image features other than the abnormal features are updated to the base image feature library associated with the target image processing. In response to the absence of abnormal features in the target image features, all the target image features are updated to the basic image feature library associated with the target image processing.

4. The method for updating the image feature library for verification according to claim 3, characterized in that, The step of determining whether there are anomalous features in the target image features includes: Calculate the cross-similarity between each of the target image features in the same candidate feature group to obtain the similarity result; Based on the similarity results, outlier analysis is used to determine whether there are any abnormal features in the target image features.

5. The method for updating the image feature library for verification according to claim 1, characterized in that, The step of selecting candidate feature groups from the feature groups based on the confidence level in the feature groups includes: Calculate the mean and standard deviation of the confidence scores in each feature group, and select at least the feature group with the largest mean or the smallest standard deviation as the candidate feature group; or Calculate the mean and standard deviation of the confidence scores in each feature group, and sum the mean and standard deviation by weight. If the weight of the mean is greater than a preset ratio, at least the feature group with the largest sum is selected as the candidate feature group. If the weight of the standard deviation is greater than the preset ratio, at least the feature group with the smallest sum is selected as the candidate feature group.

6. The method for updating the image feature library for verification according to claim 1, characterized in that, The step of selecting multiple target image features based on the confidence level in the candidate feature group includes: The candidate feature groups are sorted according to their confidence levels. The image feature vector that ranks at a specified sequence position based on confidence level is selected as the target image feature.

7. The method for updating the image feature library for verification according to claim 2, characterized in that, The step of updating the basic image feature library associated with the target image processing based on the target image features in the image feature pool includes: Obtain the historical image features stored in the basic image feature library; Cache the historical image features and clear the basic image feature library; The confidence scores of the historical image features and the target image features are sorted by size, and a preset number of image features with the highest confidence scores are selected and stored in the basic image feature library.

8. The method for updating the image feature library for verification according to claim 2, characterized in that, The step of updating the basic image feature library associated with the target image processing based on the target image features in the image feature pool further includes: In response to the fact that the remaining storage space in the basic image feature library is less than the amount of data of the target image feature to be stored, some or all of the historical image features already stored in the basic image feature library are deleted, and the target image feature to be stored is stored in the basic image feature library. In response to the remaining storage space in the basic image feature library being greater than or equal to the amount of data of the target image feature to be stored, the target image feature to be stored is stored in the basic image feature library.

9. An image feature library updating device, characterized in that, The system includes an interconnected memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method for updating the image feature library for verification as described in any one of claims 1-8.

10. A method for verifying image processing results, characterized in that, include: The image processing model is used to perform target image processing on the image to be processed to obtain the image processing result; The target image processing associated basic image feature library is obtained, and the basic image feature library is obtained by the image feature library update method for verification according to any one of claims 1-8; The image processing results are verified based on the aforementioned basic image feature library.

11. A device for verifying image processing results, characterized in that, It includes an interconnected memory and a processor, wherein the memory is used to store a computer program, which, when executed by the processor, is used to implement the image processing result verification method of claim 10.

12. A computer-readable storage medium for storing a computer program, characterized in that, When executed by a processor, the computer program is used to implement the method for updating the image feature library for verification as described in any one of claims 1-8 or the method for verifying the image processing results as described in claim 10.

Images