An image filing method, an image filing apparatus, and a computer-readable storage medium

Abstract

The application discloses an image clustering method, an image clustering device and a computer readable storage medium. The method comprises the following steps: obtaining a clustering archive of different to-be-clustered targets, wherein the clustering archive comprises to-be-clustered images; determining that the clustering archive is a suspected error archive after detecting that the clustering archive meets a preset determination condition; inputting the suspected error archive into an error identification model to obtain an identification result, wherein the identification result comprises a determination result of whether the to-be-clustered targets in suspected error images are the same, and the suspected error images are to-be-clustered images in the suspected error archive; and determining whether to remove the suspected error images from the clustering archive based on the identification result to form a new archive. In this way, the correctness of clustering can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an image archiving method, an image archiving device and a computer readable storage medium. BACKGROUND

[0002] With the development of face recognition technology, face recognition is applied in various industries, generating a large amount of trajectory data of people. The similarity of images in the trajectory data can be calculated, similar images can be clustered, and images belonging to a person can be integrated as much as possible. However, since the archiving threshold in the clustering process is generally constant, it is possible that images belonging to B will be clustered into the archive of A, resulting in an error archive. The error archive is not conducive to the development of subsequent applications, and therefore it is necessary to locate the error images in the error archive. SUMMARY

[0003] The present application provides an image archiving method, an image archiving device and a computer readable storage medium, which can improve the correctness of archiving.

[0004] To solve the above technical problems, the technical solution adopted by the present application is to provide an image archiving method, which comprises: obtaining a clustering archive of different to-be-archived targets, the clustering archive comprising to-be-archived images; determining that the clustering archive is a suspected error archive after detecting that the clustering archive meets a preset determination condition; inputting the suspected error archive into an error recognition model to obtain a recognition result, the recognition result comprising a determination result of whether the to-be-archived targets in suspected error images are the same, the suspected error images being to-be-archived images in the suspected error archive; and determining whether to remove the suspected error images from the clustering archive based on the recognition result to form a new archive.

[0005] To solve the above technical problems, another technical solution adopted by the present application is to provide an image archiving device, which comprises a memory and a processor connected to each other, wherein the memory is used to store a computer program, and the computer program is used to implement the image archiving method in the above technical solution when executed by the processor.

[0006] To solve the above technical problems, another technical solution adopted by the present application is to provide a computer readable storage medium, which is used to store a computer program, and the computer program is used to implement the image archiving method in the above technical solution when executed by the processor.

[0007] By the above scheme, the beneficial effects of the present application are: firstly, a clustering archive including multiple images to be clustered is obtained; then it is judged whether the clustering archive meets a preset judging condition, if yes, the clustering archive is recorded as a suspected error archive, the suspected error archive being an archive of images possibly including at least two clustering targets; then the suspected error archive is processed by using an error identification model to obtain a discrimination result of whether the clustering targets in each suspected error image are identical; then it is determined according to the discrimination result whether the suspected error image needs to be removed from the clustering archive, if yes, a new archive is formed; if no, it is indicated that the clustering is correct, and no new archive is formed; the error identification model can automatically discriminate whether there is an error image in the suspected error archive, and remove the error image to form a new archive when there is an error image, thereby ensuring the correctness of the clustering, and since the suspected error archive is firstly discriminated and then the error image is discriminated, the speed of the overall judgment is accelerated. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort. Among them:

[0009] Figure 1 is a flowchart of an embodiment of the image clustering method provided by the present application;

[0010] Figure 2 is a flowchart of another embodiment of the image clustering method provided by the present application;

[0011] Figure 3 is a structural diagram of an embodiment of the image clustering device provided by the present application;

[0012] Figure 4 is a structural diagram of an embodiment of the computer readable storage medium provided by the present application. DETAILED DESCRIPTION

[0013] The present application will be further described in detail below with reference to the drawings and embodiments. It is particularly pointed out that the following embodiments are only used to illustrate the present application, but do not limit the scope of the present application. Similarly, the following embodiments are only some embodiments of the present application, but not all embodiments of the present application, and all other embodiments obtained by those skilled in the art without any creative effort are within the scope of protection of the present application.

[0014] Reference to an "embodiment" in this application means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. It is expressly understood that the embodiments described herein are merely examples from among a great variety of embodiments that can be made in light of the present disclosure.

[0015] It should be noted that the terms "first", "second", "third", etc., are used herein only to describe various conditions, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second", "third" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, such as two, three, etc., unless otherwise specifically limited. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0016] First, the technical terms involved in the present application are explained.

[0017] Portrait clustering: calculating the similarity between images, and clustering images with high similarity into an archive, which is generally considered that all images in the archive belong to one person.

[0018] Error archive: an archive containing images of two or more persons at the same time. Generally, the images of the same person in the error archive are referred to as correct images, and the images of other persons are referred to as error images.

[0019] Archive splitting: locating error images in error archives by rules or machine learning techniques, and removing or splitting error images from error archives into new archives.

[0020] In the related art, there are solutions to split error images, but only compare portrait images with archive feature values to identify error images. Generally, the archive feature value is the average feature value of all images in the archive or the feature value of the preferred representative image in the archive. Therefore, such error image comparison and identification is one-sided, and a threshold set too high may result in error images not being removed, and a threshold set too low may result in correct images being removed from the archive, because the intra-class similarity difference between the same person in the portrait image clustering may be greater than the inter-class difference between different persons.

[0021] Based on the above problems, the present application provides an image clustering scheme, mainly related to the security field and the image processing field, especially related to portrait recognition clustering and data mining, which uses deep learning technology to identify error images in the clustering archives, and splits the archives containing error images. The technical scheme provided by the present application will be described in detail below.

[0022] Please refer to Figure 1 , Figure 1 is a flowchart of an embodiment of the image clustering method provided by the present application, which comprises:

[0023] Step 11: Obtain the clustering archives of different target images to be clustered.

[0024] The clustering archives include images to be clustered, which are images that need to be classified, and can be portrait images (including face images or body images), which can be obtained from an image database. Alternatively, an image capturing device is used to capture images of different target images to be clustered, and the target images to be clustered are targets that need to be clustered, which can be people, other animals or objects. After obtaining a plurality of images to be clustered, a clustering method can be used to cluster the images to be clustered to obtain clustering archives, such as K-means clustering or hierarchical clustering.

[0025] Step 12: After detecting that the clustering archives meet the preset determination condition, determine that the clustering archives are suspected error archives.

[0026] After obtaining the clustering archives, it can be determined whether the images to be clustered in the clustering archives meet the preset determination condition. If the images to be clustered meet the preset determination condition, it indicates that there are at least two target images to be clustered in the clustering archives, and at this time the clustering archives are recorded as suspected error archives. Specifically, the characteristic values of the images to be clustered in the clustering archives are used to screen suspected error images (i.e. images to be clustered in suspected error archives) to form suspected error archives.

[0027] Further, the feature extraction method can be used to calculate the feature values of each image in the clustering archive, and then the similarity between the feature values of any two images is calculated, and the number of images whose similarity with other images is less than a set threshold is calculated. If the number is greater than a set number, it indicates that the similarity between the image and other images is low, and it is determined that the clustering archive meets the preset determination condition, i.e., there are at least two target images in the clustering archive. Alternatively, the similarity between the feature value of each image and a set feature value can also be calculated. The set feature value can be the average of the feature values of all images in the clustering archive or the feature value of the image with the highest image quality. Then, it is determined whether the similarity is less than a set threshold. If the similarity is less than the set threshold, it indicates that the similarity between the image and other images is low, and it is determined that the clustering archive meets the preset determination condition.

[0028] Step 13: inputting the suspected error archive into the error recognition model to obtain a recognition result.

[0029] The error recognition model can be a graph neural network model, such as a graph convolutional network (GCN), a graph attention network, a graph autoencoder, a graph generative network, or a graph spatial-temporal network. Training data including images of at least two target images can be obtained in advance. By inputting the training data into the error recognition model and continuously adjusting the parameters of the model, a trained graph neural network model can be obtained. It can be understood that the training of the error recognition model is similar to the method of model training in related technologies, which will not be described in detail here.

[0030] After receiving the suspected error archive, the error recognition model analyzes and processes the suspected error archive to generate a recognition result. The recognition result includes a discrimination result of whether the target images in the suspected error images are the same, which is used to indicate whether each suspected error image is a correct image or an error image.

[0031] Step 14: based on the recognition result, determining whether to remove the suspected error images from the clustering archive to form a new archive.

[0032] After the recognition result is obtained, if the determination result of each suspected misfiled image in the recognition result is a correct image, it indicates that there is only one target to be clustered in the clustering archive; if the determination result of a suspected misfiled image in the recognition result is an incorrect image, and the determination result of some suspected misfiled images is a correct image, it indicates that there are at least two targets to be clustered in the clustering archive, at this time, a new archive needs to be created, and the incorrect images are moved from the clustering archive to the new archive; then the new archive can be analyzed and processed as described above to determine whether there are different images of the target to be clustered in the new archive, if so, the images are removed, and a new archive is created again, until the new archive created only contains one target to be clustered.

[0033] The embodiment provides a method for screening abnormal points in a clustering archive based on a graph neural network model, uses the graph neural network model to automatically determine incorrect images in a suspected incorrect archive, removes the incorrect images and forms a new archive, does not need to pre-set a threshold value, can automatically determine incorrect images and attribute the incorrect images to the new archive, and ensures the correctness of clustering.

[0034] Please refer to Figure 2 , Figure 2 is a flowchart of another embodiment of an image clustering method provided by the application, and the technical scheme adopted by the embodiment is described by taking a person as an example, the method comprises the following steps:

[0035] Step 21: performing clustering processing on the images to be clustered to obtain a clustering archive.

[0036] The trajectory sequence data of different targets to be clustered is obtained, and the trajectory sequence data comprises feature values of the images to be clustered, and capture times or capture locations of the images to be clustered. For example, the trajectory sequence data of all persons in a time range and a space range can be obtained, for example, the trajectory (A1, A2, A3, A4, A5) generated by the activity of person A on the same day, A1-A5 comprises information such as capture times, capture regions and image features.

[0037] After the trajectory sequence data is obtained, the clustering processing can be performed based on the feature values of the images to be clustered to obtain a clustering archive; specifically, the similarity between the feature values of the images to be clustered and the feature values corresponding to each existing clustering archive is calculated, and the maximum value of all similarities is calculated; it is judged whether the maximum value of the similarity is greater than a second similarity threshold value; if the maximum value of the similarity is greater than the second similarity threshold value, the images to be clustered are attributed to the clustering archive corresponding to the maximum value of the similarity; if the maximum value of the similarity is less than / equal to the second similarity threshold value, the images to be clustered are attributed to a new clustering archive.

[0038] In a specific embodiment, the personnel trajectory image features can be compared with the existing archive features in chronological order of personnel activities, for example, the image features of trajectory A1 of personnel A are compared with the features of all archives, if the similarity between the image features of trajectory A1 and the features of archive B meets the pre-set threshold (i.e. greater than the second similarity threshold), the image corresponding to trajectory A1 is classified into archive B; if all archives are less than / equal to the second similarity threshold, a new archive C is created, and the image corresponding to trajectory A1 is classified into archive C.

[0039] It can be understood that if there is no archive currently, a new archive D is directly established, and the image corresponding to trajectory A1 is classified into archive D; or if the similarity between the features of multiple archives and the image features of trajectory A1 is greater than the second similarity threshold, for example: archives E-F, and the similarity between the features of archive E and the image features of trajectory A1 is greater than the similarity between the features of archive F and the image features of trajectory A1, the image corresponding to trajectory A1 is classified into archive E.

[0040] Step 22: Obtain the feature value of each image to be clustered in the clustering archive, and calculate the average value of the feature values of all images to be clustered in the clustering archive to obtain the average feature value.

[0041] After obtaining the clustering archive, a suspected error archive can be located; specifically, after the images to be clustered on the action trajectories of all personnel are attributed to the corresponding clustering archives, the feature values of all images to be clustered in the clustering archive and the average feature value of the clustering archive are calculated.

[0042] Step 23: Determine whether the similarity between the feature value of each image to be clustered in the clustering archive and the average feature value is greater than a third similarity threshold.

[0043] After obtaining the feature value, the similarity between each image to be clustered in the clustering archive and the average feature value of the clustering archive is calculated according to the following formula:

[0044]

[0045] Wherein, A0 is the average feature value of the clustering archive A, B0 is the feature value of a certain image to be clustered in the clustering archive A, S is the similarity, the value of S is [-1, 1], the greater the value of S, the more similar A0 and B0 are. It can be understood that the similarity can also be calculated using Manhattan distance, and the similarity is not limited to being calculated using cosine similarity.

[0046] After obtaining the similarity between each image to be clustered in the clustering archive and the average feature value of the clustering archive, each similarity is compared with a pre-set similarity threshold (i.e., a third similarity threshold) to determine whether the clustering archive is a suspected error archive.

[0047] Step 24: If the similarity between the feature value of each image to be clustered in the clustering archive and the average feature value is not all greater than the third similarity threshold, the clustering archive is a suspected error image archive.

[0048] If there is a similarity between a certain image to be clustered in the clustering archive and the average feature value of the clustering archive that is less than or equal to the third similarity threshold, the clustering archive is determined to be a suspected error archive, i.e., there may be at least two images of people in the archive. Repeating steps 22-24 above, all suspected error archives in all clustering archives can be obtained.

[0049] Step 25: Generating a distance matrix based on the suspected error archive.

[0050] The distance matrix includes the similarity between each suspected error image in the suspected error archive and the remaining images. The absolute value of the difference between the feature values of any two suspected error images can be calculated first; then the absolute value is summed to obtain the similarity (denoted as L1 similarity), i.e., the L1 similarity is calculated using the following formula:

[0051] L1 = sum(abs(E1-E2))

[0052] Where E1 is the feature value of the suspected error image I1, and E2 is the feature value of the suspected error image I2.

[0053] The L1 similarity between all suspected error images in the suspected error archive is calculated to form a distance matrix (or can be referred to as a distance graph). Each suspected error image is a node of the graph, e.g., node H1 represents image H1, node H2 represents image H2, and node H3 represents image H3. The nodes are connected by edges, and the weight of the edge is the L1 similarity between image H1 and image H2, as shown below:

[0054]

[0055] Where the L1 similarity between image H1 / H2 / H3 and itself is constant at 0, and the greater the L1 similarity, the less similar it is.

[0056] Step 26: Inputting the suspected error archive and the distance matrix into an error recognition model to obtain a recognition result.

[0057] The following scheme can be used to identify and locate the error images in the suspected error archive:

[0058] (1) Manually label the real error archives.

[0059] The label value of the error image in the error archive is 1, and the label value of the correct image is 0. The labeled error archive is used as a training sample of the graph neural network.

[0060] (2) An error archive is input as a training sample of the error identification model.

[0061] Each training sample contains a set of feature vectors of each image in the error archive, each feature vector corresponds to a node, a distance matrix and a binary vector label to indicate whether the node is an error image.

[0062] (3) Train the graph neural network.

[0063] The graph neural network model used is deep GCN, and binary cross entropy can be used as the loss function. The training goal is to minimize the value of the loss function. The trained model theoretically has the ability to judge whether the node in the graph is an error node (i.e. an error image), that is, the label of the error node is marked as 0. The trained model is saved to facilitate loading and reuse when predicting error images in unknown archives.

[0064] (4) Input the located suspected error archive to the trained error identification model to identify error images.

[0065] The identification result output by the error identification model includes the label value corresponding to the suspected error image; specifically, the suspected error image in each suspected error archive is marked 0 or 1, indicating whether the suspected error image is an error image. If the suspected error image is marked 0, it means it is an error image; if the suspected error image is marked 1, it means it is a correct image.

[0066] Step 27: Record the suspected error archive as the current to-be-processed archive.

[0067] Step 28: Determine whether the label values of all suspected error images are the same.

[0068] Determine whether the label values of all suspected error images in the current to-be-processed archive are the same; if the label values of all suspected error images are the same, determine that the to-be-processed target in all suspected error images in the current to-be-processed archive is the same, and do not perform rejection processing on the current to-be-processed archive.

[0069] Step 29: If the label values of all suspected error images are not the same, split the current to-be-processed archive to obtain at least one new archive.

[0070] First, the number of suspected misfiled images with a label value of a preset value, i.e., 0, is counted to obtain an error statistic value. Then, the current to-be-processed archive is processed based on the error statistic value to obtain a new archive.

[0071] In a specific embodiment, the error images, i.e., the suspected misfiled images with a label value of the preset value, can be removed from the current to-be-processed archive. To split the suspected misfiled archive into a new archive, the following cases can be considered:

[0072] (1) When the error statistic value is a first preset number, the error images are added to the new archive.

[0073] The first preset number is 1, i.e., when there is one error image in the current to-be-processed archive, the error image is removed from the current to-be-processed archive and added to the new archive.

[0074] (2) When the error statistic value is a second preset number, if the similarity between the error images is greater than / equal to a first similarity threshold, the error images are added to the new archive; if the similarity between the error images is less than the first similarity threshold, each error image is placed in a different new archive.

[0075] The second preset number is greater than the first preset number, and the second preset number is 2, i.e., if the number of error images is two, the similarity between the two error images is calculated, if the similarity is lower than the first similarity threshold, the two error images are split to form new archives, respectively; if the similarity is greater than / equal to the first similarity threshold, the two error images are considered to belong to the same to-be-clustered archive target, and the two error images are placed in the same new archive.

[0076] (3) When the error statistic value is a third preset number, all error images are grouped into the current to-be-processed archive, the current to-be-processed archive is input into the error recognition model, and the step of determining whether the label values of all suspected misfiled images are the same is returned until the error statistic value is the first preset number or the second preset number.

[0077] The third preset number is greater than the second preset number, if the number of error images is greater than two, all error images are first attributed to a temporary archive, the temporary archive is recorded as the current to-be-processed archive, then the current to-be-processed archive is input into the trained deep GCN model to obtain a current recognition result, and step 27 is returned to be executed until the error statistic value is 1 or 2 to obtain a final clustering result.

[0078] The embodiment compares the similarity between any two suspected error files through establishing the image connected graph in the suspected error file, adds more information, and enhances the accuracy of the discrimination compared with the related art which only compares the feature value of the image with the feature value of the file. Moreover, the deep GCN model in the deep learning is established, the suspected error file is automatically discriminated by using the deep GCN model, the adjacent relationship of each suspected error file in the suspected error file is constructed, the information of the connected graph can be further extracted, the error image is more accurately and quickly recognized, and the speed of the overall judgment is accelerated.

[0079] Referring to Figure 3 , Figure 3 is a structural schematic diagram of an embodiment of an image clustering device provided by the application. The image clustering device 30 comprises a memory 31 and a processor 32 connected to each other. The memory 31 is configured to store a computer program. The computer program, when executed by the processor 32, is configured to implement the image clustering method in the above embodiment.

[0080] Referring to Figure 4 , Figure 4 is a structural schematic diagram of an embodiment of a computer readable storage medium provided by the application. The computer readable storage medium 40 is configured to store a computer program 41. The computer program 41, when executed by a processor, is configured to implement the image clustering method in the above embodiment.

[0081] The computer readable storage medium 40 can be a server, a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0082] In the several embodiments provided by the application, it should be understood that the disclosed method and device can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the modules or units is only a logical function division. There can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0083] The units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e. they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, some or all of the units can be selected to achieve the purpose of the embodiment.

[0084] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0085] The above is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation using the content of the present application specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. An image aggregation method, characterized in that, include: Obtain clustering archives for different target files to be clustered, wherein the clustering archives include images of the target files to be clustered; After detecting that the clustered file meets the preset judgment conditions, the clustered file is determined to be a suspected erroneous file; The suspected erroneous file is input into the error recognition model to obtain the recognition result. The recognition result includes the judgment result of whether the target of the file to be aggregated in the suspected erroneous file image is the same. The suspected erroneous file image is the file to be aggregated in the suspected erroneous file. The recognition result includes the tag value corresponding to the suspected erroneous file image. Based on the identification results, determining whether to remove the suspected misclassified image from the clustered archives to form a new archive includes: Record the suspected erroneous files as currently pending files; Determine whether the tag values ​​of all the suspected misfiled images are the same; If not, the current file to be processed is split to obtain at least one new file; The step of splitting the current file to be processed to obtain at least one new file includes: counting the number of suspected error images marked with a preset value among all the suspected error images to obtain an error statistics value; and splitting the current file to be processed based on the error statistics value to obtain the new file, including: Remove erroneous images from the current files to be processed; the erroneous images are suspected erroneous files whose marker values ​​are the preset values. When the error statistics value is a first preset number, the error image is added to the new file; When the error statistics value is a second preset number, if the similarity between the error images is greater than or equal to a first similarity threshold, then the error images are added to the new file; if the similarity between the error images is less than the first similarity threshold, then each error image is placed in a different new file. When the error statistics value is the third preset number, all the error images are combined into the current file to be processed, the current file to be processed is input into the error recognition model, and the step of judging whether the tag values ​​of all the suspected error files are the same is returned until the error statistics value is the first preset number or the second preset number. Wherein, the first preset quantity is less than the second preset quantity, and the second preset quantity is less than the third preset quantity.

2. The image aggregation method according to claim 1, characterized in that, The step of inputting the suspected error file into the error identification model to obtain the identification result includes: A distance matrix is ​​generated based on the suspected error files, and the distance matrix includes the similarity between each suspected error file image in the suspected error files and the remaining images; The suspected error file and the distance matrix are input into the error recognition model to obtain the recognition result.

3. The image aggregation method according to claim 2, characterized in that, The step of generating a distance matrix based on the suspected erroneous files includes: Calculate the absolute value of the difference between the feature values ​​of any two of the suspected misfiled images; The similarity is obtained by summing the absolute values.

4. The image aggregation method according to claim 1, characterized in that, The method further includes: If the tag values ​​of all the suspected misfiled images are the same, it is determined that the target of the file to be aggregated in all the suspected misfiled images in the current file to be processed is the same, and the current file to be processed is not removed.

5. The image aggregation method according to claim 1, characterized in that, The step of obtaining clustering archives for different target clustering documents includes: The images to be clustered are subjected to clustering processing to obtain the clustered archives.

6. The image aggregation method according to claim 5, characterized in that, The step of performing clustering processing on the image to be clustered to obtain the clustered archive includes: Acquire trajectory sequence data of the target file to be aggregated within a preset time and a preset activity range, wherein the trajectory sequence data includes feature values ​​of the image of the target file to be aggregated; Clustering is performed on the feature values ​​of the image to be clustered to obtain the clustered archive.

7. The image aggregation method according to claim 6, characterized in that, The step of performing clustering processing based on the feature values ​​of the image to be clustered to obtain the clustered archive includes: Calculate the similarity between the feature values ​​of the image to be clustered and the feature values ​​corresponding to each existing clustered file, and calculate the maximum value of all similarities; Determine whether the maximum value of the similarity is greater than the second similarity threshold; If so, the image to be clustered is assigned to the cluster file corresponding to the maximum value of the similarity. If not, the image to be clustered will be assigned to a new cluster file.

8. The image aggregation method according to claim 1, characterized in that, Before the step of determining that the clustered file is a suspected erroneous file after detecting that the clustered file meets the preset judgment conditions, the following steps are included: Obtain the feature value of each image in the clustering archive; Calculate the average feature value of all the images to be clustered in the clustered archive to obtain the average feature value; Determine whether the similarity between the feature value of each image to be clustered in the clustered archive and the average feature value is greater than a third similarity threshold; If not, then the clustered archives are determined to meet the preset judgment conditions.

9. An image aggregation device, characterized in that, The system 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 aggregation method according to any one of claims 1-8.

10. 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 image aggregation method according to any one of claims 1-8.

Images