Image aggregation method, device, nonvolatile storage medium, and electronic device

Abstract

The application discloses an image clustering method and device, a nonvolatile storage medium and an electronic device. The method comprises: obtaining a preset image set, wherein the preset image set comprises a plurality of high-quality images and a plurality of low-quality images; clustering the preset image set to obtain a plurality of image clustering sets, wherein each image clustering set comprises: a main clustering image randomly selected from the preset image set, a first slave clustering image with a high similarity to the main clustering image higher than a first similarity threshold, and a second slave clustering image with a high similarity to the main clustering image higher than a second similarity threshold; clustering the plurality of image clustering sets to obtain at least one image joint set; and establishing a corresponding image archive for each image joint set. The application solves the technical problem of low clustering accuracy caused by the fact that the existing clustering technology cannot utilize low-quality images.

Description

Technical Field

[0001] This invention relates to the field of image archiving, and more specifically, to an image archiving method, apparatus, non-volatile storage medium, and electronic device. Background Technology

[0002] With the widespread use of cameras, surveillance systems can capture massive amounts of video data. To quickly obtain behavioral information for each person from this vast amount of video data, individuals appearing in the video data can be grouped into a single file. This allows multiple images of the same person to be aggregated into that person's profile, and by analyzing that profile, their behavior can be quickly analyzed.

[0003] In the context of large-scale facial data capture and the extraction of its value, organizing and classifying facial data, clustering facial images by feature, and forming a series of facial profiles can accelerate facial retrieval, generate facial activity trajectories, and enhance the practical value of facial data in security.

[0004] However, existing face clustering technology has high computational complexity, requires a lot of hardware resources, and feature clustering can only be performed based on high-quality images, making it impossible to effectively utilize low-quality captured images, thus affecting the clustering accuracy.

[0005] There is currently no effective solution to the problem of low aggregation accuracy caused by the inability of existing aggregation technology to effectively utilize low-quality images. Summary of the Invention

[0006] The present invention provides an image aggregation method, apparatus, non-volatile storage medium, and electronic device to at least solve the technical problem of low aggregation accuracy caused by the inability of existing aggregation technology to utilize low-quality images.

[0007] According to one aspect of the present invention, an image clustering method is provided, comprising: acquiring a preset image set, wherein the preset image set includes: a plurality of high-quality images and a plurality of low-quality images; clustering the preset image set to obtain a plurality of image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image whose similarity to the master cluster image is higher than a first similarity threshold, and the second secondary cluster image... The image is a low-quality image whose similarity to the main cluster image is higher than a second similarity threshold; multiple image cluster sets are clustered to obtain at least one image joint set, wherein each image joint set includes: one or more of the image cluster sets, and when the image joint set includes multiple image cluster sets, at least one of the main cluster images in the image joint set has a similarity to the target cluster image higher than a third similarity threshold, and the target cluster image is a main cluster image randomly selected from the image joint set; a corresponding image file is established for each image joint set.

[0008] Optionally, clustering the preset image set to obtain multiple image cluster sets includes: clustering a high-quality image set to obtain multiple pre-cluster sets, wherein the high-quality image set includes multiple high-quality images in the preset image set, and each pre-cluster set includes the main cluster image and the first secondary cluster image; determining the second secondary cluster image of each main cluster image in the low-quality image set to obtain multiple image cluster sets, wherein the low-quality image set includes multiple low-quality images in the preset image set.

[0009] Optionally, clustering a high-quality image set to obtain multiple pre-cluster sets includes: randomly selecting a high-quality image from the high-quality image set as the main cluster image; traversing the high-quality image set to determine a first secondary cluster image of the main cluster image to obtain the pre-cluster set, wherein, after determining the first secondary cluster image, the pre-cluster set is removed from the high-quality image set, and the pre-cluster set is determined again.

[0010] Optionally, traversing the high-quality image set to determine the first sub-cluster image of the main cluster image and obtaining the pre-cluster set includes: determining the similarity between each high-quality image in the high-quality image set and the main cluster image; selecting high-quality images in the high-quality image set with a similarity greater than the first similarity threshold as the first sub-cluster image of the main cluster image; and determining the pre-cluster set based on the main cluster image and the first sub-cluster image.

[0011] Optionally, determining the second sub-cluster image of each of the main cluster images in the low-quality image set to obtain multiple image cluster sets includes: placing the main cluster images of the multiple pre-cluster sets into a main image set; randomly selecting the main cluster image in the main image set as the target main cluster image; traversing the low-quality image set to determine the second sub-cluster image of the target main cluster image to obtain the image cluster set, wherein after determining the second sub-cluster image of the target main cluster image, the target main cluster image is removed from the main image set, and the image cluster set is determined again.

[0012] Optionally, traversing the low-quality image set to determine the second sub-cluster images of the target main cluster image to obtain the image cluster set includes: determining the similarity between each low-quality image in the low-quality image set and the target main cluster image; selecting low-quality images in the low-quality image set with a similarity greater than the second similarity threshold as the second sub-cluster images of the target main cluster image; and placing the second sub-cluster images into a pre-cluster set of the target cluster image to obtain the image cluster set, wherein, after determining the image cluster set, low-quality images already placed into the image cluster set are removed from the low-quality image set.

[0013] Optionally, clustering multiple image cluster sets to obtain at least one image joint set includes: placing the master cluster images of the multiple image cluster sets into the joint image set; sorting the multiple master cluster images in the joint image set according to the order in which the master cluster images were selected to obtain the joint image queue; determining the similarity of adjacent master cluster images in the joint image queue according to the sorting order; and clustering adjacent master cluster images into the same image joint set if the similarity of adjacent master cluster images is greater than the third similarity threshold.

[0014] Optionally, establishing a corresponding image file for each image union set includes: selecting a representative image for each image union set according to a preset rule, wherein the representative image is selected from the main cluster image and the first secondary cluster image of the image union set; determining the matching degree between the representative image and each sample image in a preset retrieval library, wherein the preset retrieval library includes: a plurality of sample images and a real-name file corresponding to each sample image; determining the sample image with the highest matching degree in the preset retrieval library as the matching image of the representative image; if the matching degree between the matching image and the representative image is higher than a preset matching degree threshold, matching the real-name file corresponding to the matching image with the representative image, wherein the image file includes the real-name file.

[0015] According to another aspect of the present invention, an image clustering apparatus is also provided, comprising: an acquisition module, configured to acquire a preset image set, wherein the preset image set includes: a plurality of high-quality images and a plurality of low-quality images; and a first clustering module, configured to cluster the preset image set to obtain a plurality of image cluster sets, wherein each image cluster set includes: a master clustering image, a first secondary clustering image, and a second secondary clustering image, wherein the master clustering image is a high-quality image randomly selected from the preset image set, the first secondary clustering image is a high-quality image whose similarity to the master clustering image is higher than a first similarity threshold, and the second secondary clustering image is a high-quality image whose similarity to the master clustering image is higher than a first similarity threshold. The first image is a low-quality image whose similarity to the main cluster image is higher than a second similarity threshold. The second clustering module is used to cluster multiple image cluster sets to obtain at least one image joint set, wherein each image joint set includes one or more image cluster sets. When the image joint set includes multiple image cluster sets, at least one main cluster image in the image joint set has a similarity to the target cluster image higher than a third similarity threshold. The target cluster image is a main cluster image randomly selected from the image joint set. The second image establishment module is used to establish a corresponding image file for each image joint set.

[0016] According to another aspect of the present invention, a non-volatile storage medium is also provided, the non-volatile storage medium being used to store a program, wherein, when the program is running, the device where the non-volatile storage medium is located is controlled to execute the above-described image aggregation method.

[0017] According to another aspect of the present invention, an electronic device is also provided, including: a memory and a processor, the processor being configured to run a program stored in the processor, wherein the program executes the above-described image aggregation method when it runs.

[0018] In this embodiment of the invention, a preset image set is obtained, comprising: multiple high-quality images and multiple low-quality images; the preset image set is clustered to obtain multiple image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the master cluster image, and the second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the master cluster image; the multiple image cluster sets are clustered to obtain at least one image joint set, wherein each image joint set includes: one or more image cluster sets. When the image union set includes multiple image cluster sets, at least one main cluster image in the image union set has a similarity higher than the third similarity threshold with the target cluster image. The target cluster image is a main cluster image randomly selected from the image union set. By establishing a corresponding image archive for each image union set, the purpose of performing three-level threshold clustering with quality control on the preset image set is achieved. This allows the image union set generated by clustering to cover both low-quality and high-quality images. Thus, by establishing an image archive based on the image union set, the image archive can cover both low-quality and high-quality images, thereby improving the technical effect of clustering accuracy. This solves the technical problem of low clustering accuracy caused by the inability of existing clustering techniques to utilize low-quality images. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0020] Figure 1 This is a flowchart of an image aggregation method according to an embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of a first-level clustering according to an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of a second-level clustering according to an embodiment of the present invention;

[0023] Figure 4 This is a schematic diagram of a third-level clustering according to an embodiment of the present invention;

[0024] Figure 5 This is a schematic diagram of a face data aggregation process for one person, one file, according to an embodiment of the present invention;

[0025] Figure 6 This is a schematic diagram of a clustering process with quality control according to an embodiment of the present invention;

[0026] Figure 7 This is a schematic diagram of an image aggregation device according to an embodiment of the present invention;

[0027] Figure 8 This is a structural block diagram of a computer terminal according to an embodiment of the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] According to an embodiment of the present invention, an image aggregation method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] Figure 1 This is a flowchart of an image aggregation method according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0032] Step S102: Obtain a preset image set, wherein the preset image set includes: multiple high-quality images and multiple low-quality images;

[0033] Step S104: Cluster the preset image set to obtain multiple image cluster sets. Each image cluster set includes: a main cluster image, a first secondary cluster image, and a second secondary cluster image. The main cluster image is a high-quality image randomly selected from the preset image set. The first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the main cluster image. The second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the main cluster image.

[0034] Step S106: Cluster multiple image cluster sets to obtain at least one image joint set, wherein each image joint set includes: one or more image cluster sets. When the image joint set includes multiple image cluster sets, at least one main cluster image in the image joint set has a similarity higher than a third similarity threshold with a target cluster image. The target cluster image is a main cluster image randomly selected from the image joint set.

[0035] Step S108: Create a corresponding image file for each image union set.

[0036] In this embodiment of the invention, a preset image set is obtained, comprising: multiple high-quality images and multiple low-quality images; the preset image set is clustered to obtain multiple image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the master cluster image, and the second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the master cluster image; the multiple image cluster sets are clustered to obtain at least one image joint set, wherein each image joint set includes: one or more image cluster sets. When the image union set includes multiple image cluster sets, at least one main cluster image in the image union set has a similarity higher than the third similarity threshold with the target cluster image. The target cluster image is a main cluster image randomly selected from the image union set. By establishing a corresponding image archive for each image union set, the purpose of performing three-level threshold clustering with quality control on the preset image set is achieved. This allows the image union set generated by clustering to cover both low-quality and high-quality images. Thus, by establishing an image archive based on the image union set, the image archive can cover both low-quality and high-quality images, thereby improving the technical effect of clustering accuracy. This solves the technical problem of low clustering accuracy caused by the inability of existing clustering techniques to utilize low-quality images.

[0037] In step S102 above, the preset image set may include multiple captured images. After the captured images are collected, the captured images can be scored for image quality. Capture images with image quality scores higher than a preset quality threshold are identified as high-quality images, and capture images with image quality scores not higher than the preset quality threshold are identified as low-quality images.

[0038] Optionally, the captured image can be a facial image, which can be captured by a camera.

[0039] Optionally, the captured images from the camera can be used to extract feature vectors through a deep neural network to obtain a face image.

[0040] Optionally, the captured images are scored for image quality, mainly taking into account factors such as face size, blur level, light intensity, head posture angle, and degree of occlusion.

[0041] In step S104 above, since the preset image set includes multiple high-quality images and multiple low-quality images, clustering the preset image set is equivalent to clustering the high-quality images and low-quality images in the preset image set.

[0042] Optionally, during the clustering process, a master clustering image can be selected from the high-quality images, and then high-quality images with a similarity higher than a first similarity threshold to the master clustering image can be used as the first secondary clustering images of the master clustering image, thereby achieving clustering of high-quality images.

[0043] Optionally, after clustering high-quality images, since the main cluster image has been determined, by calculating the similarity between low-quality images and the main cluster image, low-quality images with a similarity higher than the second similarity threshold with the main cluster image can be used as the second sub-cluster images of the main cluster image, thereby realizing the clustering of low-quality images.

[0044] Optionally, the first similarity threshold is less than the second similarity threshold.

[0045] Optionally, each image cluster set has a corresponding master cluster image, which serves as the cluster center of the corresponding image cluster set. Each image cluster set may have any number of first secondary cluster images and second secondary cluster images.

[0046] In step S106 above, the image union set can be the clustering result of multiple principal clustering images. Since each image clustering set has a corresponding principal clustering image, after clustering the principal clustering image, the image clustering set containing the principal clustering image can be synchronously placed into the image union set containing the principal clustering image, following the clustering result of the principal clustering image. Thus, each image union set can include at least one image clustering set.

[0047] In step S108 above, each image union set includes at least one image cluster set. Each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image. Therefore, each image union set can include multiple high-quality images and low-quality images, and high-quality images and low-quality images in the same image union set can have the same characteristics. For example, in the case where the high-quality images and low-quality images are face images, each image union set can include multiple high-quality images and low-quality images of the same person.

[0048] Optionally, each image set may have the same characteristics, and an image archive may be constructed based on these same characteristics. For example, an image archive may represent multiple live images of the same person, and the live images may include high-quality images and low-quality images of the same person.

[0049] As an optional embodiment, clustering a preset image set to obtain multiple image cluster sets includes: clustering a high-quality image set to obtain multiple pre-cluster sets, wherein the high-quality image set includes multiple high-quality images in the preset image set, and each pre-cluster set includes a master cluster image and a first sub-cluster image; determining a second sub-cluster image for each master cluster image in a low-quality image set to obtain multiple image cluster sets, wherein the low-quality image set includes multiple low-quality images in the preset image set.

[0050] In the above embodiments of the present invention, during the process of clustering a preset image set, high-quality images in the image clustering set can be clustered first to obtain a pre-clustering set; then, low-quality images in the image clustering set can be assigned to the pre-clustering set to obtain multiple image clustering sets including high-quality images and low-quality images, thereby realizing quality-based clustering of the preset image set.

[0051] Optionally, the preset image set can be divided into a high-quality image set and a low-quality image set according to image quality, and then clustering can be performed based on the divided high-quality image set and low-quality image set.

[0052] As an optional embodiment, clustering a high-quality image set to obtain multiple pre-cluster sets includes: randomly selecting a high-quality image from the high-quality image set as the main cluster image; traversing the high-quality image set to determine the first secondary cluster image of the main cluster image to obtain the pre-cluster set, wherein, after determining the first secondary cluster image, the pre-cluster set is removed from the high-quality image set, and the pre-cluster set is determined again.

[0053] In the above embodiments of the present invention, during the clustering of high-quality images, a high-quality image can be randomly selected from the high-quality image set as the master clustering image, and the selected master clustering image is used as the leader point (i.e., as the cluster center) to establish a pre-cluster set. Then, the high-quality image set is traversed until all the first sub-cluster images of the master clustering image have been placed into the pre-cluster set. Then, the high-quality images that have been placed into the pre-cluster set are removed from the high-quality image set. Then, a new master clustering image is selected to establish the next pre-cluster set. This process continues until all the high-quality images in the high-quality image set are assigned to the pre-cluster set, thereby achieving the clustering of high-quality images.

[0054] Figure 2 This is a schematic diagram of a first-level clustering according to an embodiment of the present invention, as shown below. Figure 2 As shown, the first-level clustering is used for high-quality image clustering. Solid dots represent the main cluster images, hollow dots represent the first secondary cluster images, numbers represent the selection order of the main cluster images, and large circles represent the scope of each main cluster image. Hollow dots within the large circle represent the first secondary cluster images of that main cluster image. The first secondary cluster image, even if it is within the radius of multiple main cluster images, belongs only to the main cluster image with the smaller index.

[0055] As an optional embodiment, traversing the high-quality image set to determine the first sub-cluster image of the main cluster image and obtaining the pre-cluster set includes: determining the similarity between each high-quality image in the high-quality image set and the main cluster image; selecting high-quality images in the high-quality image set whose similarity is greater than a first similarity threshold as the first sub-cluster image of the main cluster image; and determining the pre-cluster set based on the main cluster image and the first sub-cluster image.

[0056] In the above embodiments of the present invention, the selected main clustering image is used as the leader point (i.e., as the cluster center), and the images with a similarity higher than the first similarity threshold from the remaining high-quality images are selected as the first secondary clustering images of the main clustering image (i.e., the subject points of the leader point), thus establishing a pre-clustering set and realizing the generation of the pre-clustering set.

[0057] As an optional embodiment, determining the second sub-cluster image of each master cluster image in the low-quality image set to obtain multiple image cluster sets includes: placing the master cluster images of multiple pre-cluster sets into a master image set; randomly selecting a master cluster image from the master image set as the target master cluster image; traversing the low-quality image set to determine the second sub-cluster images of the target master cluster image to obtain the image cluster set, wherein, after determining the second sub-cluster images of the target master cluster image, the target master cluster image is removed from the master image set, and the image cluster set is determined again.

[0058] In the above embodiments of the present invention, each pre-cluster set contains a master cluster image. During the clustering of low-quality images, the master cluster image can be extracted from multiple pre-cluster sets and placed into the master image set. Then, a master cluster image is randomly selected from the master image set and used as the leader point (i.e., the cluster center). The second sub-cluster image of the master cluster image is selected from the low-quality image set and placed into the pre-cluster set corresponding to the master cluster image to generate an image cluster set. Then, the low-quality image set is traversed until all the second sub-cluster images of the master cluster image have been placed into the pre-cluster set. Then, the low-quality images that have been placed into the pre-cluster set are removed from the low-quality image set. Then, the master cluster image is reselected to establish the next image cluster set. This process continues until the remaining low-quality images in the low-quality image set cannot be used as the second sub-cluster images of any master cluster image, thereby achieving the clustering of low-quality images.

[0059] Optionally, if a low-quality image in a low-quality image set cannot be assigned to any pre-cluster set, the low-quality image is discarded.

[0060] Figure 3 This is a schematic diagram of a second-level clustering according to an embodiment of the present invention, as shown below. Figure 3 As shown, the second-level clustering is used to cluster low-quality images. Solid dots represent primary cluster images, hollow dots represent low-quality images, numbers indicate the selection order of primary cluster images, and large circles represent the scope of each primary cluster image. Hollow dots within the large circle represent the second sub-cluster images of that primary cluster image. Even if a second sub-cluster image falls within the radius of multiple primary cluster images, it only belongs to the primary cluster image with the smaller index. Hollow dots not within any large circle represent discarded low-quality images.

[0061] As an optional embodiment, traversing the low-quality image set to determine the second sub-cluster images of the target main cluster image and obtaining the image cluster set includes: determining the similarity between each low-quality image in the low-quality image set and the target main cluster image; selecting low-quality images in the low-quality image set whose similarity is greater than a second similarity threshold as the second sub-cluster images of the target main cluster image; and placing the second sub-cluster images into the pre-cluster set of the target cluster image to obtain the image cluster set, wherein, after determining the image cluster set, low-quality images that have been placed into the image cluster set are removed from the low-quality image set.

[0062] In the above embodiments of the present invention, the selected main clustering image is used as the leader point (i.e., as the cluster center). Low-quality images with a similarity higher than the second similarity threshold to the main clustering image are selected from the low-quality image set and used as the second subordinate clustering images (i.e., the subject points of the leader point) of the main clustering image. The second subordinate clustering images are then placed into the pre-clustering set corresponding to the main clustering image to generate an image clustering set, thus realizing the generation of the image clustering set.

[0063] As an optional embodiment, clustering multiple image cluster sets to obtain at least one image joint set includes: placing the main cluster images of the multiple image cluster sets into the joint image set; sorting the multiple main cluster images in the joint image set according to the order in which the main cluster images were selected, to obtain the joint image queue; determining the similarity of adjacent main cluster images in the joint image queue in the order of arrangement; and clustering adjacent main cluster images into the same image joint set if the similarity of adjacent main cluster images is greater than a third similarity threshold.

[0064] In the above embodiments of the present invention, the process of clustering multiple image cluster sets can be transformed into clustering the main cluster image in each image cluster set, and then, based on the clustering results of the main cluster image, synchronously dividing the image cluster set into the joint image set where the corresponding main cluster image is located, thereby completing the clustering of multiple image cluster sets.

[0065] Figure 4 This is a schematic diagram of a third-level clustering according to an embodiment of the present invention, as shown below. Figure 4 As shown, the third-level clustering is used to cluster multiple image cluster sets. Solid dots represent the main cluster images, and the dashed large circles represent the scope of the main cluster images. When two main cluster images fall within each other's scope, a connecting line will be generated. The connected graph obtained by connecting the main cluster images is a joint image set.

[0066] As an optional embodiment, establishing a corresponding image file for each image union set includes: selecting a representative image for each image union set according to preset rules, wherein the representative image is selected from the main cluster image and the first sub-cluster image of the image union set; determining the matching degree between the representative image and each sample image in a preset retrieval library, wherein the preset retrieval library includes: multiple sample images and a real-name file corresponding to each sample image; determining the sample image with the highest matching degree in the preset retrieval library as the matching image of the representative image; if the matching degree between the matching image and the representative image is higher than a preset matching degree threshold, matching the real-name file corresponding to the matching image with the representative image, wherein the image file includes the real-name file.

[0067] In the above embodiments of the present invention, a corresponding image file is established for each image union set. The image union set can be searched against pre-set sample images in a preset retrieval library to determine the sample image that matches the image union set. Then, the pre-set real-name file of the sample image is matched with the image union set to realize the establishment of an image file for the image union set. Furthermore, in order to simplify the retrieval process of the image union set, a representative image that can represent the characteristics of the image union set can be selected from the high-quality images of each image union set. This representative image replaces the image union set in the preset retrieval library to determine the real-name file that matches the image union set.

[0068] Alternatively, the representative image can be the principal cluster image from the image union set.

[0069] Optionally, if no sample image matching the representative image exists in the preset search database, an anonymous profile can be created for the representative image, wherein the image profile also includes a real-name profile.

[0070] As an optional example, the preset search library can be pre-built using multiple high-quality images after deduplication. By clustering multiple high-quality images, a sample cluster set can be obtained. The sample cluster set includes a main sample image and a secondary sample image. The secondary sample image is a high-quality image whose similarity to the main sample image is higher than a first similarity threshold. Then, multiple sample cluster sets are clustered to obtain at least one sample joint set. When the sample joint set includes multiple sample cluster sets, at least one main sample image in the sample joint set has a similarity to the target sample image higher than a third similarity threshold. The target sample image is a main sample image randomly selected from the sample joint set. Then, sample images are selected from each sample joint set, and real-name profiles are created.

[0071] As an optional example, before determining the second cluster image, each low-quality image in the preset image set can be searched in a preset retrieval library to determine the matching degree between the low-quality image and each sample image in the preset retrieval library. The sample image with the highest matching degree in the preset retrieval library is determined as the representative image. If the matching degree between the matching image and the low-quality image is higher than a preset matching degree threshold, the real-name file corresponding to the matching image is matched with the low-quality image. If the matching degree between the matching image and the low-quality image is not higher than the preset matching degree threshold, it is determined whether the low-quality image belongs to the second cluster image.

[0072] This invention also provides a preferred embodiment, which offers a method for creating individual face archives, reducing computational complexity, improving archive accuracy, reducing hardware resources, and enhancing overall product competitiveness. The archive method of this invention takes a face image as input and outputs the archive to which the face image belongs. In the preprocessing stage, a deep neural network is used to extract feature vectors and quality scores from the face images, and the captured images are divided into high-quality and low-quality images based on a quality threshold. The individual-file archive process consists of five stages: the first stage is deduplication through clustering in the database; the second stage is retrieving low-quality images from the database; the third stage is quality-based clustering of the captured images; the fourth stage is retrieving representative images for each cluster from the database; and the fifth stage is generating real-name and anonymous archives.

[0073] It should be noted that the base database is the preset search database, and the captured images are high-quality and low-quality images in the preset image set.

[0074] Figure 5 This is a schematic diagram of a face data aggregation process for one person, one file, according to an embodiment of the present invention, as shown below. Figure 5 As shown, it includes the following stages:

[0075] The first stage involves clustering and deduplication of the base database graphs. If the graphs in the base database have unique identifiers, deduplication can be performed using these identifiers. Otherwise, the quality-based clustering method used in the third stage can be employed. The features of the deduplicated graphs are then added to the database to construct the retrieval database.

[0076] In the second stage, low-quality captured images are searched in the database. A match is considered found if the highest score of the search result exceeds a threshold; otherwise, it is considered a non-match.

[0077] In the third stage, high-quality snapshots and low-quality snapshots that did not match in the second stage are clustered by quality.

[0078] In the fourth stage, a representative graph for each class is retrieved from the base database. Using the cluster representative graph generated in the fourth stage, a search is conducted in the base database. If the highest score retrieved exceeds a threshold, a match is considered found; otherwise, a match is considered not found.

[0079] In the fifth stage, named and anonymous profiles are generated. All images in the class corresponding to the low-quality snapshots matched in the second stage and the representative images matched in the fourth stage are grouped into named profiles corresponding to the matched base database images. The classes corresponding to the representative images not matched in the fourth stage are grouped into anonymous profiles.

[0080] Figure 6 This is a schematic diagram of a clustering process with quality control according to an embodiment of the present invention, such as... Figure 6As shown, the first-level clustering clusters high-quality images based on a first similarity threshold, dividing them into primary cluster images and primary secondary cluster images. The second-level clustering clusters low-quality images based on a second similarity threshold and the primary cluster images, dividing them into secondary secondary cluster images and discarded images. The third-level clustering uses a third similarity threshold to cluster the primary cluster images into a joint image set. Finally, the clusters are sorted and representative images are selected. The specific process is as follows:

[0081] S61, the first-level clustering includes:

[0082] S611, divide all high-quality images into two sets: the processed image set and the unprocessed image set. Initially, all high-quality images belong to the unprocessed image set.

[0083] S612, randomly select an image from the unprocessed image set, which is called the principal clustering image (e.g., leader point);

[0084] S613, calculate its similarity to all remaining unprocessed images;

[0085] S614, for each unprocessed image, if its similarity with the main cluster image exceeds a first similarity threshold, then this unprocessed image is classified as the first sub-cluster image (such as a high-quality subject point) of the main cluster image.

[0086] S615, move both the main clustering image and the first secondary clustering image from the unprocessed image set to the processed image set;

[0087] S616, Repeat S612-S615 until all images are returned to the processed image set.

[0088] S62, the second-level clustering includes:

[0089] S621, all low-quality images are divided into two sets: the processed image set and the unprocessed image set. Initially, all images belong to the unprocessed image set.

[0090] S622, extract one principal cluster image (e.g., leader point) in sequence according to the order of principal cluster images;

[0091] S623, calculate its similarity to all remaining unprocessed images;

[0092] S624, for each unprocessed image, if its similarity to the main cluster image exceeds a second similarity threshold, then this unprocessed image is classified as a second sub-cluster image (such as a low-quality subject point) of the main cluster image.

[0093] S625, the second clustered image is moved from the set of unprocessed images to the set of processed images;

[0094] S626, repeat S622-S625 until all main cluster images have been processed. Low-quality images that do not belong to any main cluster image are called discarded images.

[0095] S63, third-level clustering, clusters the main cluster images into image union sets (such as lord unions), including:

[0096] S631, assign a union ID to each of the main cluster images, and the initial union ID of the i-th main cluster image is i;

[0097] S632, Traverse each main cluster image in ascending order of its index, and calculate the similarity between the main cluster image and all main cluster images with higher indices.

[0098] S633, if the similarity between the main cluster image i and the main cluster image j exceeds the third similarity threshold and the union ids of the main cluster images i and j are different, then the unions to which the two main cluster images belong are merged; the merging rule is that for a union with fewer main cluster images, the union ids of all its main cluster images are changed to the union ids of the unions with more main cluster images.

[0099] S634, repeat S632-S633 until all main cluster images have been processed.

[0100] S64, Organizing the clustering results and selecting representative images for each class includes:

[0101] S641, for each image union set (such as a lord union), include all its master cluster images, as well as the corresponding first and second sub-cluster images, into this union;

[0102] S642, select one of the high-quality images from all the high-quality images in each image union set as the representative image of the image union set, according to the rule that the high-quality image is closest to the average value of all the high-quality images in the image union set.

[0103] S643, renumber the final union IDs starting from 0, and assign a new union ID to each image as its cluster number.

[0104] As an optional example, the method of creating a facial recognition file for each individual includes the following steps:

[0105] The first step is to preprocess all base images and captured images to obtain quality scores for all images, classifying them into high-quality images and low-quality images.

[0106] The second step is to remove duplicate images from the base database. The original base database contains approximately 1 million high-quality, clear frontal images. A clustering program was run using a V100 graphics card to remove duplicates, resulting in approximately 420,000 images. The base database was then constructed for retrieval, taking about 15 minutes.

[0107] The third step involves searching the database for the top 1 low-quality captured images. There are approximately 5 million low-quality captured images in total, with about 2 million matching and about 3 million not matching. The search process is executed on the V100 graphics card and takes about 30 minutes.

[0108] The fourth step involves combining high-quality capture images with unmatched low-quality capture images for quality-controlled clustering. There are approximately 10 million high-quality capture images and approximately 3 million unmatched low-quality capture images, totaling 13 million images. These images are clustered into approximately 1.3 million classes. The clustering process is executed on a V100 graphics card and takes approximately 100 minutes.

[0109] The fifth step involves retrieving the top-ranked representative graph features for each class from the base database. This retrieval process is executed on a V100 graphics card and takes approximately 10 minutes. About 700,000 classes were matched, while approximately 600,000 classes were not.

[0110] The sixth step is to merge the low-quality snapshots matched in the third step with all the images corresponding to the class matched in the fifth step to form real-name files. After merging, there are approximately 320,000 real-name files, totaling about 10 million images.

[0111] The seventh step involves creating anonymous files from the classes matched in the fifth step that contain at least two images, resulting in approximately 200,000 anonymous files and a total of approximately 2 million images.

[0112] According to an embodiment of the present invention, an image aggregation device embodiment is also provided. It should be noted that the image aggregation device can be used to execute the image aggregation method in the embodiment of the present invention, and the image aggregation method in the embodiment of the present invention can be executed in the image aggregation device.

[0113] Figure 7 This is a schematic diagram of an image aggregation device according to an embodiment of the present invention, such as... Figure 7As shown, the device may include: an acquisition module 72, used to acquire a preset image set, wherein the preset image set includes: multiple high-quality images and multiple low-quality images; and a first clustering module 74, used to cluster the preset image set to obtain multiple image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image whose similarity to the master cluster image is higher than a first similarity threshold, and the second secondary cluster image is a high-quality image whose similarity to the master cluster image is higher than a first similarity threshold. The first image is a low-quality image whose similarity to the second image is higher than a second similarity threshold. The second clustering module 76 is used to cluster multiple image cluster sets to obtain at least one image joint set. Each image joint set includes one or more image cluster sets. When the image joint set includes multiple image cluster sets, at least one main cluster image in the image joint set has a similarity to the target cluster image that is higher than a third similarity threshold. The target cluster image is a main cluster image randomly selected from the image joint set. The second image establishment module 78 is used to establish a corresponding image file for each image joint set.

[0114] It should be noted that the acquisition module 72 in this embodiment can be used to execute step S102 in this application embodiment, the first clustering module 74 in this embodiment can be used to execute step S104 in this application embodiment, the second clustering module 76 in this embodiment can be used to execute step S106 in this application embodiment, and the establishment module 78 in this embodiment can be used to execute step S108 in this application embodiment. The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments.

[0115] In this embodiment of the invention, a preset image set is obtained, comprising: multiple high-quality images and multiple low-quality images; the preset image set is clustered to obtain multiple image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the master cluster image, and the second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the master cluster image; the multiple image cluster sets are clustered to obtain at least one image joint set, wherein each image joint set includes: one or more image cluster sets. When the image union set includes multiple image cluster sets, at least one main cluster image in the image union set has a similarity higher than the third similarity threshold with the target cluster image. The target cluster image is a main cluster image randomly selected from the image union set. By establishing a corresponding image archive for each image union set, the purpose of performing three-level threshold clustering with quality control on the preset image set is achieved. This allows the image union set generated by clustering to cover both low-quality and high-quality images. Thus, by establishing an image archive based on the image union set, the image archive can cover both low-quality and high-quality images, thereby improving the technical effect of clustering accuracy. This solves the technical problem of low clustering accuracy caused by the inability of existing clustering techniques to utilize low-quality images.

[0116] As an optional embodiment, the first clustering module includes: a first clustering unit, used to cluster a high-quality image set to obtain multiple pre-cluster sets, wherein the high-quality image set includes multiple high-quality images in a preset image set, and each pre-cluster set includes a master cluster image and a first secondary cluster image; and a second clustering unit, used to determine a second secondary cluster image for each master cluster image in a low-quality image set to obtain multiple image cluster sets, wherein the low-quality image set includes multiple low-quality images in a preset image set.

[0117] As an optional embodiment, the first clustering unit includes: a first selection unit, configured to randomly select a high-quality image from the high-quality image set as the main clustering image; and a first determination unit, configured to traverse the high-quality image set, determine a first secondary clustering image of the main clustering image, and obtain a pre-clustering set, wherein, after determining the first secondary clustering image, the pre-clustering set is removed from the high-quality image set, and the pre-clustering set is determined again.

[0118] As an optional embodiment, the first determining unit includes: a first determining subunit, used to determine the similarity between each high-quality image in the high-quality image set and the main clustering image; a second determining subunit, used to select high-quality images in the high-quality image set whose similarity is greater than a first similarity threshold as the first sub-clustering images of the main clustering image; and a third determining subunit, used to determine a pre-clustering set based on the main clustering image and the first sub-clustering image.

[0119] As an optional embodiment, the second clustering unit includes: a preprocessing unit for placing the master clustering images of multiple pre-clustering sets into a master image set; a second selection unit for randomly selecting a master clustering image from the master image set as the target master clustering image; and a second determination unit for traversing the low-quality image set to determine the second sub-clustering images of the target master clustering image, thereby obtaining an image clustering set, wherein after determining the second sub-clustering images of the target master clustering image, the target master clustering image is removed from the master image set, and the image clustering set is determined again.

[0120] As an optional embodiment, the second determining unit includes: a fourth determining subunit, used to determine the similarity between each low-quality image in the low-quality image set and the target main clustering image; a fifth determining subunit, used to select low-quality images in the low-quality image set whose similarity is greater than a second similarity threshold as second sub-clustering images of the target main clustering image; and a sixth determining subunit, used to put the second sub-clustering images into a pre-clustering set of the target clustering image to obtain an image clustering set, wherein, after determining the image clustering set, low-quality images already put into the image clustering set are removed from the low-quality image set.

[0121] As an optional embodiment, the second clustering module includes: an insertion unit for inserting the main cluster images of multiple image clustering sets into a joint image set; a sorting unit for sorting the multiple main cluster images in the joint image set according to the order in which the main cluster images were selected, to obtain the joint image queue; a third determination unit for determining the similarity of adjacent main cluster images in the joint image queue in the order of arrangement; and a third clustering unit for clustering adjacent main cluster images into the same image joint set when the similarity of adjacent main cluster images is greater than a third similarity threshold.

[0122] As an optional embodiment, the establishment module includes: a third selection unit, used to select a representative image of each image union set according to preset rules, wherein the representative image is selected from the main cluster image and the first sub-cluster image of the image union set; a fourth determination unit, used to determine the matching degree between the representative image and each sample image in a preset retrieval library, wherein the preset retrieval library includes: multiple sample images and a real-name file corresponding to each sample image; a fifth determination unit, used to determine the sample image with the highest matching degree in the preset retrieval library as the matching image of the representative image; and a matching unit, used to match the real-name file corresponding to the matching image with the representative image when the matching degree between the matching image and the representative image is higher than a preset matching degree threshold, wherein the image file includes a real-name file.

[0123] Embodiments of the present invention can provide a computer terminal, which can be any computer terminal device in a group of computer terminals. Optionally, in this embodiment, the computer terminal can also be replaced by a mobile terminal or other terminal device.

[0124] Optionally, in this embodiment, the computer terminal may be located in at least one of a plurality of network devices in a computer network.

[0125] In this embodiment, the computer terminal described above can execute the program code for the following steps in the image clustering method: obtaining a preset image set, wherein the preset image set includes: multiple high-quality images and multiple low-quality images; clustering the preset image set to obtain multiple image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold with the master cluster image, and the second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold with the master cluster image; clustering the multiple image cluster sets to obtain at least one image joint set, wherein each image joint set includes: one or more image cluster sets, wherein when the image joint set includes multiple image cluster sets, at least one master cluster image in the image joint set has a similarity higher than a third similarity threshold with a target cluster image, and the target cluster image is a master cluster image randomly selected from the image joint set; and establishing a corresponding image file for each image joint set.

[0126] Optionally, Figure 8 This is a structural block diagram of a computer terminal according to an embodiment of the present invention. Figure 8 As shown, the computer terminal 8 may include one or more (only one is shown in the figure) processors 82 and memory 84.

[0127] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the image aggregation method and apparatus in this embodiment of the invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned image aggregation method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal 80 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0128] The processor can invoke information and application programs stored in memory via a transmission device to perform the following steps: acquiring a preset image set, wherein the preset image set includes: multiple high-quality images and multiple low-quality images; clustering the preset image set to obtain multiple image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the master cluster image, and the second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the master cluster image; clustering the multiple image cluster sets to obtain at least one image joint set, wherein each image joint set includes: one or more image cluster sets, wherein when the image joint set includes multiple image cluster sets, at least one master cluster image in the image joint set has a similarity higher than a third similarity threshold to a target cluster image, and the target cluster image is a master cluster image randomly selected from the image joint set; and establishing a corresponding image file for each image joint set.

[0129] Optionally, the processor may also execute program code for the following steps: clustering a high-quality image set to obtain multiple pre-cluster sets, wherein the high-quality image set includes multiple high-quality images in a preset image set, and each pre-cluster set includes a master cluster image and a first sub-cluster image; determining a second sub-cluster image for each master cluster image in a low-quality image set to obtain multiple image cluster sets, wherein the low-quality image set includes multiple low-quality images in a preset image set.

[0130] Optionally, the processor may also execute program code that performs the following steps: randomly selects a high-quality image from the high-quality image set as the master clustering image; traverses the high-quality image set, determines the first sub-clustering image of the master clustering image, and obtains a pre-clustering set, wherein, after determining the first sub-clustering image, the pre-clustering set is removed from the high-quality image set, and the pre-clustering set is determined again.

[0131] Optionally, the processor may also execute program code for the following steps: determining the similarity between each high-quality image in the high-quality image set and the main clustering image; selecting high-quality images in the high-quality image set whose similarity is greater than a first similarity threshold as the first sub-clustering images of the main clustering image; and determining a pre-clustering set based on the main clustering image and the first sub-clustering images.

[0132] Optionally, the processor may also execute program code that performs the following steps: placing the master clustering image of multiple pre-clustering sets into a master image set; randomly selecting a master clustering image from the master image set as the target master clustering image; traversing the low-quality image set to determine the second sub-clustering image of the target master clustering image, thereby obtaining an image clustering set, wherein, after determining the second sub-clustering image of the target master clustering image, the target master clustering image is removed from the master image set, and the image clustering set is determined again.

[0133] Optionally, the processor may also execute program code for the following steps: determining the similarity between each low-quality image in the low-quality image set and the target main cluster image; selecting low-quality images in the low-quality image set whose similarity is greater than a second similarity threshold as second sub-cluster images of the target main cluster image; and placing the second sub-cluster images into the pre-cluster set of the target cluster image to obtain an image cluster set, wherein, after determining the image cluster set, low-quality images that have been placed into the image cluster set are removed from the low-quality image set.

[0134] Optionally, the processor may also execute program code for the following steps: placing the principal cluster images of multiple image cluster sets into a joint image set; sorting the principal cluster images in the joint image set according to the order in which they were selected, to obtain the joint image queue; determining the similarity of adjacent principal cluster images in the joint image queue in the order of arrangement; and clustering adjacent principal cluster images into the same image joint set if the similarity of adjacent principal cluster images is greater than a third similarity threshold.

[0135] Optionally, the processor may also execute program code for the following steps: selecting a representative image from each image union set according to preset rules, wherein the representative image is selected from the main cluster image and the first sub-cluster image of the image union set; determining the matching degree between the representative image and each sample image in a preset retrieval library, wherein the preset retrieval library includes: multiple sample images and real-name files corresponding to each sample image; determining the sample image with the highest matching degree in the preset retrieval library as the matching image of the representative image; if the matching degree between the matching image and the representative image is higher than a preset matching degree threshold, matching the real-name file corresponding to the matching image with the representative image, wherein the image file includes the real-name file.

[0136] This invention provides an image clustering scheme. In this embodiment, a preset image set is obtained, comprising multiple high-quality images and multiple low-quality images. The preset image set is then clustered to obtain multiple image cluster sets. Each image cluster set includes a master cluster image, a first secondary cluster image, and a second secondary cluster image. The master cluster image is a high-quality image randomly selected from the preset image set. The first secondary cluster images are high-quality images whose similarity to the master cluster image is higher than a first similarity threshold. The second secondary cluster images are low-quality images whose similarity to the master cluster image is higher than a second similarity threshold. The multiple image cluster sets are then clustered to obtain at least one image joint set. Each image joint set includes one or more image cluster sets. When the image union set includes multiple image cluster sets, at least one main cluster image in the image union set has a similarity higher than the third similarity threshold with the target cluster image. The target cluster image is a main cluster image randomly selected from the image union set. By establishing a corresponding image archive for each image union set, the purpose of performing three-level threshold clustering with quality control on the preset image set is achieved. This allows the image union set generated by clustering to cover both low-quality and high-quality images. Thus, by establishing an image archive based on the image union set, the image archive can cover both low-quality and high-quality images, thereby improving the technical effect of clustering accuracy. This solves the technical problem of low clustering accuracy caused by the inability of existing clustering techniques to utilize low-quality images.

[0137] Those skilled in the art will understand that Figure 8 The structure shown is for illustrative purposes only. The computer terminal can also be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a mobile internet device (MID), a PAD, and other terminal devices. Figure 8 This does not limit the structure of the aforementioned electronic device. For example, the computer terminal 80 may also include components that are more advanced than those described above. Figure 8The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 8 The different configurations shown.

[0138] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a non-volatile medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0139] Embodiments of the present invention also provide a non-volatile storage medium. Optionally, in this embodiment, the aforementioned non-volatile storage medium can be used to store the program code executed by the image archiving method provided in the above embodiments.

[0140] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0141] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: obtaining a preset image set, wherein the preset image set includes: multiple high-quality images and multiple low-quality images; clustering the preset image set to obtain multiple image cluster sets, wherein each image cluster set includes: a master cluster image, a first secondary cluster image, and a second secondary cluster image, wherein the master cluster image is a high-quality image randomly selected from the preset image set, the first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the master cluster image, and the second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the master cluster image; clustering the multiple image cluster sets to obtain at least one image joint set, wherein each image joint set includes: one or more image cluster sets, wherein when the image joint set includes multiple image cluster sets, at least one master cluster image in the image joint set has a similarity higher than a third similarity threshold to a target cluster image, and the target cluster image is a master cluster image randomly selected from the image joint set; and establishing a corresponding image file for each image joint set.

[0142] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: clustering a high-quality image set to obtain multiple pre-cluster sets, wherein the high-quality image set includes multiple high-quality images in a preset image set, and each pre-cluster set includes a master cluster image and a first slave cluster image; determining a second slave cluster image for each master cluster image in a low-quality image set to obtain multiple image cluster sets, wherein the low-quality image set includes multiple low-quality images in the preset image set.

[0143] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: randomly selecting a high-quality image from the high-quality image set as the main clustering image; traversing the high-quality image set to determine the first secondary clustering image of the main clustering image to obtain a pre-clustering set, wherein, after determining the first secondary clustering image, the pre-clustering set is removed from the high-quality image set, and the pre-clustering set is determined again.

[0144] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: determining the similarity between each high-quality image in the high-quality image set and the main clustering image; selecting high-quality images in the high-quality image set whose similarity is greater than a first similarity threshold as the first sub-clustering images of the main clustering image; and determining a pre-clustering set based on the main clustering image and the first sub-clustering images.

[0145] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: placing the master cluster image of multiple pre-cluster sets into a master image set; randomly selecting a master cluster image from the master image set as the target master cluster image; traversing the low-quality image set to determine the second sub-cluster image of the target master cluster image, thereby obtaining an image cluster set, wherein, after determining the second sub-cluster image of the target master cluster image, the target master cluster image is removed from the master image set, and the image cluster set is determined again.

[0146] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: determining the similarity between each low-quality image in the low-quality image set and the target master cluster image; selecting low-quality images in the low-quality image set whose similarity is greater than a second similarity threshold as second sub-cluster images of the target master cluster image; placing the second sub-cluster images into the pre-cluster set of the target cluster image to obtain an image cluster set, wherein, after determining the image cluster set, low-quality images that have been placed into the image cluster set are removed from the low-quality image set.

[0147] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: placing the main cluster images of multiple image cluster sets into a joint image set; sorting the multiple main cluster images in the joint image set according to the order in which the main cluster images were selected, to obtain the joint image queue; determining the similarity of adjacent main cluster images in the joint image queue in the order of arrangement; and clustering adjacent main cluster images into the same image joint set if the similarity of adjacent main cluster images is greater than a third similarity threshold.

[0148] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: selecting a representative image for each image union set according to a preset rule, wherein the representative image is selected from the main cluster image and the first sub-cluster image of the image union set; determining the matching degree between the representative image and each sample image in a preset retrieval library, wherein the preset retrieval library includes: multiple sample images and a real-name file corresponding to each sample image; determining the sample image with the highest matching degree in the preset retrieval library as the matching image of the representative image; if the matching degree between the matching image and the representative image is higher than a preset matching degree threshold, matching the real-name file corresponding to the matching image with the representative image, wherein the image file includes a real-name file.

[0149] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0150] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0151] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0152] 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0153] Furthermore, the functional units in the various embodiments of the present invention 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.

[0154] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a non-volatile storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0155] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An image aggregation method, characterized in that, include: Obtain a preset image set, wherein the preset image set includes: multiple high-quality images and multiple low-quality images; The preset image set is clustered to obtain multiple image cluster sets, wherein each image cluster set includes: a main cluster image, a first secondary cluster image, and a second secondary cluster image. The main cluster image is a high-quality image randomly selected from the preset image set. The first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the main cluster image. The second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the main cluster image. Clustering multiple image cluster sets yields at least one image joint set, wherein each image joint set includes one or more image cluster sets. When the image joint set includes multiple image cluster sets, at least one main cluster image in the image joint set has a similarity higher than a third similarity threshold with a target cluster image. The target cluster image is a main cluster image randomly selected from the image joint set. Create a corresponding image archive for each of the aforementioned image union sets.

2. The method according to claim 1, characterized in that, Clustering the preset image set yields multiple image cluster sets, including: Clustering of a high-quality image set yields multiple pre-cluster sets, wherein the high-quality image set includes multiple high-quality images from the preset image set, and each pre-cluster set includes the main cluster image and the first secondary cluster image; In the low-quality image set, a second sub-cluster image is determined for each of the main cluster images to obtain a plurality of image cluster sets, wherein the low-quality image set includes a plurality of the low-quality images in the preset image set.

3. The method according to claim 2, characterized in that, Clustering of high-quality image sets yields several pre-cluster sets, including: The high-quality image is randomly selected from the set of high-quality images as the main clustering image; Traverse the high-quality image set to determine the first sub-cluster image of the main cluster image, and obtain the pre-cluster set. After determining the first sub-cluster image, remove the pre-cluster set from the high-quality image set, and determine the pre-cluster set again.

4. The method according to claim 3, characterized in that, Traversing the high-quality image set, determining the first sub-cluster image of the main cluster image, and obtaining the pre-cluster set includes: Determine the similarity between each high-quality image in the high-quality image set and the main cluster image; High-quality images with a similarity greater than the first similarity threshold are selected from the set of high-quality images and used as the first sub-cluster images of the main clustering image; The pre-cluster set is determined based on the main cluster image and the first secondary cluster image.

5. The method according to claim 2, characterized in that, Determining a second sub-cluster image for each of the main cluster images in a low-quality image set, resulting in a plurality of image cluster sets, includes: The main cluster images of the multiple pre-cluster sets are placed into the main image set; The main clustering image is randomly selected from the main image set and used as the target main clustering image; Traverse the set of low-quality images to determine the second sub-cluster image of the target master cluster image, thereby obtaining the image cluster set. After determining the second sub-cluster image of the target master cluster image, remove the target master cluster image from the master image set and determine the image cluster set again.

6. The method according to claim 5, characterized in that, Traversing the set of low-quality images, determining the second sub-cluster images of the target main cluster image, and obtaining the image cluster set includes: Determine the similarity between each low-quality image in the low-quality image set and the target main cluster image; Low-quality images with a similarity greater than the second similarity threshold are selected from the set of low-quality images and used as the second sub-cluster images of the target main cluster image; The second sub-clustered image is placed into the pre-clustered set of the target clustered image to obtain the image cluster set, wherein, after determining the image cluster set, low-quality images that have been placed into the image cluster set are removed from the low-quality image set.

7. The method according to claim 1, characterized in that, Clustering multiple image cluster sets to obtain at least one joint image set includes: The master cluster image of multiple image clustering sets is placed into a joint image set; According to the order in which the main cluster images were selected, the multiple main cluster images in the joint image set are sorted to obtain the joint image queue; The similarity of adjacent main cluster images in the joint image queue is determined sequentially according to the order of arrangement; If the similarity between adjacent main cluster images is greater than the third similarity threshold, the adjacent main cluster images are clustered into the same image joint set.

8. The method according to claim 1, characterized in that, Creating a corresponding image archive for each of the aforementioned image union sets includes: Representative images of each image union set are selected according to preset rules, wherein the representative images are selected from the main cluster image and the first secondary cluster image of the image union set; Determine the matching degree between the representative image and each sample image in the preset retrieval library, wherein the preset retrieval library includes: multiple sample images and real-name files corresponding to each sample image; The sample image with the highest matching degree in the preset search library is determined as the matching image of the representative image; If the matching degree between the matched image and the representative image is higher than a preset matching degree threshold, the real-name file corresponding to the matched image is matched with the representative image, wherein the image file includes the real-name file.

9. An image aggregation device, characterized in that, include: The acquisition module is used to acquire a preset image set, wherein the preset image set includes: multiple high-quality images and multiple low-quality images; The first clustering module is used to cluster the preset image set to obtain multiple image cluster sets. Each image cluster set includes: a main cluster image, a first secondary cluster image, and a second secondary cluster image. The main cluster image is a high-quality image randomly selected from the preset image set. The first secondary cluster image is a high-quality image with a similarity higher than a first similarity threshold to the main cluster image. The second secondary cluster image is a low-quality image with a similarity higher than a second similarity threshold to the main cluster image. The second clustering module is used to cluster multiple image cluster sets to obtain at least one image joint set, wherein each image joint set includes one or more image cluster sets, and when the image joint set includes multiple image cluster sets, at least one main cluster image in the image joint set has a similarity higher than a third similarity threshold with a target cluster image, and the target cluster image is a main cluster image randomly selected from the image joint set; A module is established to create a corresponding image file for each of the image union sets.

10. A non-volatile storage medium, characterized in that, The non-volatile storage medium is used to store a program, wherein, when the program is running, it controls the device where the non-volatile storage medium is located to execute the image aggregation method according to any one of claims 1 to 8.

11. An electronic device, characterized in that, include: A memory and a processor, the processor being configured to run a program stored in the memory, wherein the program, when running, performs the image aggregation method according to any one of claims 1 to 8.

Images