Image clustering methods, apparatus, computer equipment and storage media

Abstract

This invention discloses an image clustering method, apparatus, computer device, and storage medium. The method involves acquiring a sample image cluster, which includes multiple sample image nodes. After performing image clustering on each sample image node to obtain an image connectivity graph, the image connectivity graph is segmented to obtain at least one image connectivity subgraph. Each image connectivity subgraph includes at least one node relationship edge connecting two sample image nodes. The shortest path value and the robustness value of the relationship edges corresponding to all node relationship edges in each image connectivity subgraph are determined. Based on the shortest path value and the robustness value of the relationship edges corresponding to each node relationship edge, a target connectivity graph corresponding to the sample image cluster is determined from the image connectivity graph. This invention improves the accuracy and precision of image clustering and segmentation.

Description

Technical Field

[0001] This invention relates to the field of cluster analysis technology, and in particular to an image clustering method, apparatus, computer device, and storage medium. Background Technology

[0002] Clustering has wide applications in fields such as business, medical diagnosis, image processing, information retrieval, and bioinformatics. Clustering can divide nodes into several clusters based on the attributes of each node in the cluster, so that the nodes in each cluster have high similarity, which is convenient for applications such as information push, product recommendation, and relationship network analysis.

[0003] In existing technologies, clustering is often performed based on the similarity between nodes, and then further clustered based on the connectivity between the nodes after the clustering. However, this method has the following shortcomings: First, clustering based on the similarity between nodes may result in some nodes being abnormally segmented, leading to low clustering accuracy. Second, clustering based on connectivity may split nodes in the same set into several smaller sets, thereby reducing the accuracy of clustering. Summary of the Invention

[0004] This invention provides an image clustering method, apparatus, computer device, and storage medium to address the problem of low accuracy and precision in existing clustering technologies.

[0005] An image clustering method, comprising:

[0006] Obtain a sample image cluster; the sample image cluster includes multiple sample image nodes;

[0007] Image clustering is performed on each of the sample image nodes to obtain an image connectivity graph;

[0008] The image connected graph is segmented to obtain at least one image connected subgraph; each image connected subgraph includes at least one node relationship edge connecting two sample image nodes;

[0009] Determine the shortest path values ​​and the robustness values ​​of the relationship edges that correspond one-to-one with all node relationship edges in each of the image connected subgraphs;

[0010] Based on the shortest path value corresponding to the edge of each node and the edge strength value, a target connected graph corresponding to the sample image cluster is determined from the image connected graph.

[0011] An image clustering device, comprising:

[0012] A cluster acquisition module is used to acquire a sample image cluster; the sample image cluster includes multiple sample image nodes.

[0013] The image clustering module is used to perform image clustering on each of the sample image nodes to obtain an image connectivity graph;

[0014] A connected graph segmentation module is used to segment the image connected graph to obtain at least one image connected subgraph; one of the image connected subgraphs includes at least one node relationship edge connecting two of the sample image nodes;

[0015] The relation edge analysis module is used to determine the shortest path value and the robustness value of the relation edge corresponding one-to-one with all node relation edges in each of the image connected subgraphs.

[0016] The connected graph segmentation module is used to determine the target connected graph corresponding to the sample image cluster from the image connected graph based on the shortest path value and the firmness value of the relationship edge corresponding to each node.

[0017] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the image clustering method described above.

[0018] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described image clustering method.

[0019] The aforementioned image clustering method, apparatus, computer device, and storage medium, after clustering sample image nodes to obtain an image connected graph, segment the image connected graph based on its connectivity to obtain image connected subgraphs; then, they determine the shortest path values ​​and edge strength values ​​corresponding to the node relationships in the image connected subgraphs; since node relationships with larger shortest path values ​​have less impact on image clustering, the image connected subgraphs can be segmented by removing these edges, thereby improving the accuracy and precision of image connected graph segmentation in image clustering. This improves the accuracy of image clustering. However, since the relationship edges between nodes are constructed based on the similarity of nodes in each sample image, there may be errors in determining the similarity of nodes in the sample image. Therefore, segmenting the connected graph of the image solely based on the shortest path value may lead to bias. To address this, we can determine the robust relationship edge values ​​of each node's relationship edges by using the out-degree and in-degree of each sample image node. This can reduce the error caused by the shortest path value in clustering segmentation. Thus, by combining the shortest path value and the robust relationship edge value to perform clustering segmentation on the connected graph of the image, we can improve the accuracy and precision of image clustering segmentation. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of an image clustering method according to an embodiment of the present invention;

[0022] Figure 2 This is a flowchart of step S20 in an image clustering method according to an embodiment of the present invention;

[0023] Figure 3 This is a flowchart of step S30 in an image clustering method according to an embodiment of the present invention;

[0024] Figure 4 This is another flowchart of step S30 in the image clustering method of one embodiment of the present invention;

[0025] Figure 5 This is a flowchart of step S40 in an image clustering method according to an embodiment of the present invention;

[0026] Figure 6 This is a schematic diagram of an image clustering device according to an embodiment of the present invention;

[0027] Figure 7 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

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

[0029] In one embodiment, such as Figure 1 As shown, an image clustering method is provided, including the following steps:

[0030] S10: Obtain the sample image cluster; the sample image cluster includes multiple sample image nodes.

[0031] Understandably, clustering analysis techniques are used in scenarios such as social network analysis and community personnel analysis. For example, in community personnel analysis, it is necessary to cluster thousands of images to determine the personnel situation within the community. Therefore, this invention takes the application scenario of image clustering as an example for clustering analysis. The sample image cluster can consist of several sample images, and each sample image in the cluster can be considered a sample image node.

[0032] Furthermore, assuming that the image clustering method of the present invention is applied to, for example, community network analysis, the sample image cluster can be replaced with a community network cluster, and a community can be regarded as a community node. The target connectivity graph corresponding to the community network cluster is determined through the following steps S20 to S40.

[0033] S20: Perform image clustering on each sample image node to obtain an image connectivity graph;

[0034] S30: Segment the image connected graph to obtain at least one image connected subgraph; an image connected subgraph includes at least one node relationship edge connecting two sample image nodes.

[0035] In essence, image clustering involves determining the similarity relationships between nodes in each sample image, and then connecting these nodes using nodes with edge connections to form a connected graph, which is a directed connected graph. Furthermore, after forming the image connected graph, the image connected graph can be segmented according to the connectivity between the sample image nodes in the image connected graph. That is, there will be multiple relatively dense sample image nodes in the image connected graph (dense is defined here as the number of node relationship edges between these sample image nodes is relatively large). There may be two relatively dense sample image nodes connected by a node relationship edge. For example, suppose there are 10,000 sample image nodes in part A connected by several node relationship edges, and 1,000 sample image nodes in part B connected by several node relationship edges. If one sample image node a in part A is connected to one sample image node b in part B by a node relationship edge, the density of part A and part B will be reduced. Therefore, the node relationship edge between sample image node a and sample image node b can be deleted to restore the density of part A and part B. In this way, by segmenting the image connected graph, at least one image connected subgraph can be obtained, and each image connected subgraph has at least one node relationship edge connecting two sample image nodes.

[0036] S40: Determine the shortest path value and the firmness value of the relationship edge corresponding one-to-one with all node relationship edges in each connected subgraph of the image.

[0037] Understandably, a connected subgraph of an image may contain several sample image nodes. There may be multiple paths from one sample image node to other sample image nodes, and each path must contain at least one node-related edge. The shortest path is the path with the fewest node-related edges. Suppose there are three paths from one sample image node to another, and the number of node-related edges in these three paths are as follows: the first path contains three node-related edges, the second path contains two node-related edges, and the third path contains four node-related edges. Then, the second path is the shortest path between these two sample image nodes. After determining the shortest path between any two sample image nodes in the connected subgraph, the number of times each node-related edge appears in all shortest paths can be determined; this is the shortest path value.

[0038] Furthermore, as stated in the above description, the image connectivity graph of this embodiment is a directed connectivity graph. Therefore, the out-degree (out-degree is the number of node relationship edges from the sample image node to other sample image nodes) and in-degree (in-degree is the number of node relationship edges from other sample image nodes to the sample image node) of each sample image node can be determined. Then, the relationship edge strength value with respect to the node relationship edge can be determined based on the out-degree and in-degree of the two sample image nodes corresponding to the node relationship edge.

[0039] S50: Based on the shortest path values ​​corresponding to the edges related to each node and the strength values ​​of the edges, determine the target connected graph corresponding to the sample image cluster from the image connected graph.

[0040] Specifically, after determining the shortest path values ​​and edge strength values ​​corresponding one-to-one with all node relationship edges in each image connected subgraph, the merging values ​​corresponding to each node relationship edge can be determined based on the shortest path values ​​and edge strength values ​​corresponding to the same node relationship edge. The image connected graph is then segmented based on these merging values, thereby determining the target connected graph corresponding to the sample image cluster. The target connected graph is the connected graph with the highest density in the sample image cluster, the highest strength between each sample image node in the target connected graph, and the most stable node relationship edges connecting any two sample image nodes. Thus, step S30 determines the shortest path values ​​and edge strength values ​​of each node relationship edge in each image connected subgraph, and then segments the image connected graph using these values, eliminating the need to perform similarity calculations on each sample image node and then conduct connectivity analysis based on the similarity calculation results. This reduces the time required for clustering and segmentation, thereby improving the efficiency of image clustering.

[0041] In this embodiment, after clustering the sample image nodes to obtain an image connected graph, the connected graph is segmented based on its connectivity to obtain an image connected subgraph. Then, the shortest path value and the robustness value of the relationship edges corresponding to the node relationships in the image connected subgraph are determined. Since node relationship edges with larger shortest path values ​​have less impact on image clustering, the image connected subgraph can be segmented by removing these edges, improving the accuracy and precision of image clustering segmentation. However, since each node relationship edge is constructed based on the similarity of the sample image nodes, there may be errors in determining the similarity. Therefore, segmenting the image connected graph solely based on the shortest path value may lead to bias. Therefore, the robustness value of the relationship edges is determined by the out-degree and in-degree of each sample image node, which can reduce the error caused by the shortest path value in clustering segmentation. Thus, combining the shortest path value and the robustness value of the relationship edges for clustering segmentation of the image connected graph can improve the accuracy and precision of image clustering segmentation.

[0042] In one embodiment, in step S20, as Figure 2 As shown, image clustering is performed on each sample image node to obtain an image connectivity graph, including:

[0043] S201: Select one sample image node from all sample image nodes as the image vertex.

[0044] Understandably, the sample image cluster includes multiple sample image nodes, and thus one sample image node can be arbitrarily selected from all the sample image nodes, and the selected sample image node is used as the image vertex.

[0045] S202: Determine the image similarity between the image vertices and all unselected sample image nodes respectively.

[0046] Intuitively, image similarity is the degree of similarity between the sample image corresponding to the vertex of an image and the sample images corresponding to all unselected sample image nodes. This image similarity can be determined by methods such as cosine similarity or structural similarity measurement.

[0047] S203: Based on image similarity, among all unselected sample image nodes, select the preset number of sample image nodes with the highest image similarity as image connection nodes corresponding to image vertices, and connect the image vertex with the image connection node corresponding to that image vertex by a directed edge to obtain node relationship edges.

[0048] Understandably, the preset number of edges is used to limit the number of node relationship edges between each sample image node and other sample image nodes. The preset number of edges can be set according to requirements. For example, the preset number of edges threshold can be set to 100, 200, etc.

[0049] Specifically, after determining the image similarity between the first image vertex and all unselected sample image nodes, the image similarities can be sorted in descending order. From all unselected sample image nodes, a predetermined number of sample image nodes with the highest similarity (i.e., the predetermined number of edges with the highest similarity) are selected as the image connection nodes corresponding to the image vertex. A directed edge is then established between the image vertex and its corresponding image connection node to obtain the node relationship edge. For example, assuming the predetermined number of edges is 100, the top 100 sample image nodes with the highest similarity are selected as the image connection nodes for that image vertex.

[0050] S204: Select a new sample image node from all sample image nodes as the new image vertex. The new sample image node is the sample image node that has not been selected from all sample image nodes.

[0051] S205: Continue to determine the node relationship edges corresponding to new image vertices until all sample image nodes have been selected.

[0052] Specifically, after connecting the image vertex with the corresponding image connection node using directed edges to obtain the node relationship edges, a new sample image node can be selected from all the sample image nodes as a new image vertex (this new image vertex is not the same sample image node as the image vertex in the above steps). The node relationship edges corresponding to the new image vertex are determined through the above steps S202-S203, and another sample image node (which is not the same sample image node as the new image vertex and the image vertex) is selected. In this way, all sample image nodes are selected, that is, each sample image node has a preset number of corresponding node relationship edges, which means that all sample image nodes in the sample image cluster have been selected.

[0053] S206: Generate a connected graph based on all the obtained node relationship edges.

[0054] Specifically, when all sample image nodes have a preset number of corresponding node relationship edges, an image connectivity graph containing all sample image nodes and the node relationship edges corresponding to each sample image node can be constructed based on all the obtained node relationship edges.

[0055] In this embodiment, by introducing a preset edge number threshold, the construction of node relationship edges for each sample image node is reduced, thereby improving the construction efficiency of the image connected graph. Furthermore, directed edge connections can be made between sample image nodes with high sample similarity, reducing the impact of node relationship edges between two sample image nodes with low similarity on the image connected graph, thus improving the accuracy of image connected graph construction.

[0056] In one embodiment, in step S40, as Figure 3 As shown, this means determining the shortest path values ​​corresponding one-to-one with all node relationships in each connected subgraph of the image, including:

[0057] S401: Determine all shortest paths in the connected subgraph of the image; a shortest path is the shortest path between any two sample image nodes in the connected subgraph of the image; a shortest path contains at least one node relationship edge.

[0058] Understandably, an image connected subgraph may contain several sample image nodes. There may be multiple paths from one sample image node to other sample image nodes, and each path must contain at least one node-related edge. The shortest path is the path with the fewest node-related edges. Suppose there are three paths from one sample image node to another, and the number of node-related edges in these three paths are as follows: the first path contains three node-related edges, the second path contains two node-related edges, and the third path contains four node-related edges. Then, the second path is the shortest path between these two sample image nodes. Therefore, after segmenting the image connected graph to obtain the image connected subgraph, the shortest connected path between any two sample image nodes in the image connected subgraph can be determined. That is, every two sample image nodes in the image connected subgraph correspond to the first shortest connected path.

[0059] S402: Record the total number of times the node relationship edge appears in all connected shortest paths as the relationship edge path value corresponding to that node relationship edge.

[0060] Specifically, after determining the shortest path between any two sample image nodes in the connected subgraph of the image, since the shortest path records the node relationship edges it contains, the total number of times the node relationship edge appears in all shortest paths can be determined, and this total number is recorded as the relationship edge path value corresponding to the node relationship edge, that is, a node relationship edge corresponds to the first relationship edge path value.

[0061] S403: Determine whether there exists at least one node relationship edge whose path value is greater than the preset shortest path edge number threshold.

[0062] Understandably, a preset shortest path edge count threshold is used to limit the number of times a node relationship edge appears in all shortest paths. For example, the preset shortest path edge count threshold can be set to 5, 10, etc. For instance, when the preset shortest path edge count threshold is set to 5, it means that each node relationship edge cannot appear more than 5 times in all shortest paths.

[0063] S404: If there is no node relationship edge with at least one relationship edge path value greater than the preset shortest path edge number threshold, then the relationship edge path value corresponding to each node relationship edge is recorded as the corresponding shortest path value.

[0064] Specifically, after recording the total number of times a node relationship edge appears in all connected shortest paths as the relationship edge path value corresponding to that node relationship edge, a preset shortest path edge count threshold is obtained. Then, the relationship edge path values ​​are compared with the preset shortest path edge count threshold. If all relationship edge path values ​​are less than or equal to the preset shortest path edge count threshold, that is, there is no relationship edge path value corresponding to a node relationship edge that is greater than the preset shortest path edge count threshold, it indicates that all node relationship edges meet the requirements of the preset shortest path edge count threshold, which reflects that the density between sample image nodes in the connected subgraph of the image is high. Therefore, the relationship edge path values ​​corresponding to each node relationship edge can be recorded as the shortest path values ​​corresponding to the node relationship edges.

[0065] S405: If there is at least one node relation edge whose path value is greater than the preset shortest path edge number threshold, then the node relation edge corresponding to the largest relation edge path value is recorded as the largest relation edge, and the largest relation edge is deleted from the connected subgraph of the image. Continue to determine the relation edge path values ​​corresponding to the node relation edges of the connected subgraph of the image after deleting the largest relation edge, until there is no node relation edge whose path value is greater than the preset shortest path edge number threshold. Record the relation edge path values ​​corresponding to each node relation edge as the corresponding shortest path values.

[0066] Specifically, after determining whether there is at least one node relation edge with a path value greater than the preset shortest path edge number threshold, if there is at least one node relation edge with a path value greater than the preset shortest path edge number threshold, the node relation edge corresponding to the largest relation edge path value is directly recorded as the largest relation edge, and the largest relation edge is deleted from the connected subgraph of the image.

[0067] Furthermore, after recording the node relationship edge corresponding to the largest relationship edge path value as the largest relationship edge and deleting the largest relationship edge from the image connected subgraph, the relationship edge path value between any two sample image nodes in the image connected subgraph after deleting the largest relationship edge is determined. It is understood that this relationship edge path value may be the same as or different from the relationship edge path value in step S402, because deleting the largest relationship edge in the image connected subgraph may affect the connectivity between some sample image nodes. For example, suppose the image connected subgraph includes two parts, one part A includes 7 sample image nodes, including one sample image node a, and the other part B includes 8 sample image nodes, including one sample image node a. The image contains a sample image node b, and these two parts are connected by a node relationship edge between sample image nodes a and b. Therefore, the path between sample image nodes in part A and sample image nodes in part B must pass through the node relationship edge between sample image nodes a and b. Assuming that the node relationship edge between sample image nodes a and b is the maximum relationship edge, after deleting this maximum relationship edge from the connected image subgraph, sample image nodes in part A will not be able to reach sample image nodes in part B. Therefore, the relationship edge path values ​​of each node relationship edge in the connected image subgraph before the maximum relationship edge is deleted may be the same as or different from the relationship edge path values ​​of each node relationship edge in the connected image subgraph after the maximum relationship edge is deleted.

[0068] Furthermore, after determining the path value of the relation edge corresponding to the node relation edge in the connected subgraph of the image after deleting the maximum relation edge, if there is still at least one node with a path value greater than the preset shortest path edge number threshold, then continue to delete the node relation edge with the maximum relation edge path value (different from the maximum relation edge mentioned above) until there is no node relation edge with a path value greater than the preset shortest path edge number threshold. The path value of the relation edge corresponding to each node relation edge is recorded as the corresponding shortest path value.

[0069] Understandably, the path values ​​of the relation edges corresponding to each node's relation edge in the connected subgraph after each deletion of the maximum relation edge will be updated accordingly. For example, suppose there are 10 shortest paths in a connected subgraph. Suppose the path value of the relation edge corresponding to one node's relation edge E is 9, the path value of the relation edge corresponding to another node's relation edge F is 8, and the path value of the relation edge corresponding to other nodes' relation edges E is 4. If the preset shortest path edge number threshold is set to 3, then node relation edge E is deleted, and the shortest path value corresponding to node relation edge E is determined to be 9. And suppose that after deleting node relation edge E, the path value of the relation edge corresponding to node relation edge F in the connected subgraph is 5, and the path values ​​of the relation edges corresponding to other nodes' relation edges are all less than or equal to 3 (such as 1, 2, 3, etc.), then node relation edge F is deleted, and the shortest path value corresponding to node relation edge F is determined to be 5 (instead of the 8 mentioned above). Suppose that after deleting node relation edge F, the path values ​​of all node relation edges in the connected subgraph are 2, then the shortest path value corresponding to the remaining node relation edges is determined to be 2. That is, if it is necessary to delete the maximum relation edge, the value of the relation edge path corresponding to the maximum relation edge is directly recorded as the value of the shortest path corresponding to the maximum relation edge, and the new relation edge path values ​​of the remaining node relation edges in the connected subgraph of the image are determined until the relation edge path values ​​of all node relation edges are less than or equal to the preset shortest path edge number threshold.

[0070] In this embodiment, by determining the shortest path value corresponding to the edge of each node relationship, and comparing it with a preset shortest path edge number threshold, the shortest path values ​​are continuously filtered out. The shortest path values ​​greater than the preset threshold are then removed, and the node relationship edge with the largest shortest path value is deleted before calculating the shortest path value for the new connected subgraph. This continuously eliminates the influence of node relationship edges with large shortest path values ​​on the segmentation of the connected subgraph, resulting in a higher density of sample image nodes in the segmented connected subgraph. This provides a foundation for subsequent clustering and segmentation of the connected graph, thereby improving the accuracy and precision of clustering and segmenting the connected graph.

[0071] In one embodiment, such as Figure 4 As shown, step S40, which involves determining the robust values ​​of the relation edges that correspond one-to-one with all node relation edges in each connected subgraph of the image, includes:

[0072] S406: Record the two sample image nodes connected by the node relationship edge as the first image node and the second image node, respectively.

[0073] Understandably, as stated in the above description, each node relationship edge connects two sample image nodes. Therefore, the two sample image nodes connected by the node relationship edge are recorded as the first image node and the second image node, respectively.

[0074] S407: Determine the first node stability coefficient of the first image node in its respective image connected subgraph, and the second node stability coefficient of the second image node in its respective image connected subgraph.

[0075] Understandably, the first node robustness coefficient can be determined based on the out-degree and in-degree of the first image node in its respective image connected subgraph; the second node robustness coefficient can be determined based on the out-degree and in-degree of the second image node in its respective image connected subgraph.

[0076] In one embodiment, determining the first node robustness coefficient of the first image node in the image connected subgraph includes:

[0077] Obtain the out-degree and in-degree of the first image node in its associated connected subgraph.

[0078] Understandably, the above description indicates that the image connected graph is a directed connected graph; therefore, each image connected subgraph is also a directed connected graph. The out-degree of the first node is the number of node relationship edges between the first image node and other sample image nodes in its respective image connected subgraph; the in-degree of the first node is the number of node relationship edges between the first image node and other sample image nodes in its respective image connected subgraph.

[0079] The sum of the out-degree and in-degree of the first node is recorded as the first sum of in-degree and out-degree. Based on the node's in-degree and the first sum of in-degree and out-degree, the first node stability coefficient of the first image node in its respective image connected subgraph is determined.

[0080] Specifically, after obtaining the first node out-degree and the first node in-degree of the first image node in its respective image connected subgraph, the sum of the first node out-degree and the first node in-degree is recorded as the first sum of out-degree and in-degree, and the quotient between the node in-degree and the first sum of out-degree and in-degree is determined as the first node stability coefficient of the first image node in its respective image connected subgraph.

[0081] Further, determining the second node robustness coefficient of the second image node in its belonging image connected subgraph includes:

[0082] Obtain the out-degree and in-degree of the second image node in its associated connected subgraph.

[0083] Understandably, the above description indicates that the image connected graph is a directed connected graph; therefore, each image connected subgraph is also a directed connected graph. The out-degree of the second node is the number of node relationship edges between the second image node and other sample image nodes in its respective image connected subgraph; the in-degree of the second node is the number of node relationship edges between the second image node and other sample image nodes in its respective image connected subgraph.

[0084] The sum of the out-degree and in-degree of the second node is recorded as the second in-degree sum. Based on the in-degree and the second in-degree sum, the second node stability coefficient of the second image node in its respective image connected subgraph is determined.

[0085] Specifically, after obtaining the second node out-degree and the second node in-degree of the second image node in its respective image connected subgraph, the sum of the second node out-degree and the second node in-degree is recorded as the second out-in-degree sum, and the quotient between the node in-degree and the second out-in-degree sum is determined as the second node stability coefficient of the second image node in its respective image connected subgraph.

[0086] S408: Determine the stability coefficient of the relation edge corresponding to the node relation edge based on the stability coefficient of the first node and the stability coefficient of the second node.

[0087] Specifically, after determining the first node firmness coefficient of the first image node in its respective image connected subgraph and the second node firmness coefficient of the second image node in its respective image connected subgraph, the absolute value of the difference between the first node firmness coefficient and the second node firmness coefficient is recorded as the relation edge firmness coefficient. The larger the relation edge firmness coefficient, the more unstable the corresponding node relation edge is; conversely, the smaller the relation edge firmness coefficient, the more stable the corresponding node relation edge is.

[0088] In this embodiment, the stability coefficient of the relationship edge corresponding to each node is determined by the out-degree and in-degree of the two sample image nodes connected by the relationship edge in their respective image connected subgraphs. The stability coefficient of the relationship edge reflects the stability of the relationship edge of each node. In this way, the stability coefficient of the relationship edge can make up for the shortcomings of the shortest path value in clustering and segmenting the image connected graph. This provides a data foundation for the subsequent steps of combining the shortest path value and the stability coefficient of the relationship edge to cluster and segment the image connected graph, thereby improving the accuracy of clustering and segmentation.

[0089] In one embodiment, such as Figure 5 As shown, in step S50, that is, determining the target connected graph corresponding to the sample image cluster from the image connected graph based on the shortest path value corresponding to the relation edges of each node and the firmness value of the relation edges, includes:

[0090] S501: Determine the merging value of the relationship edges corresponding to each node based on the shortest path value and the firmness value of the relationship edges.

[0091] In one embodiment, determining the merging value of the relationship edges corresponding to each node's relationship edges based on the shortest path value and the strength value of the relationship edges includes:

[0092] Obtain the first weight value corresponding to the shortest path value, and the second weight value corresponding to the solidity value of the relation edge.

[0093] Understandably, the first weight value and the second weight value are preset values, and the sum of the first weight value and the second weight value is 1. The first weight value or the second weight value can be set according to the sample image nodes contained in the image connected subgraph.

[0094] The product of the shortest path value and the first weight value is recorded as the first relation edge value, and the product of the relation edge strength value and the second weight value is recorded as the second relation edge value.

[0095] The sum of the values ​​of the first and second relation edges corresponding to the same node is recorded as the merged relation edge value corresponding to that node.

[0096] Specifically, after obtaining the first weight value corresponding to the shortest path value and the second weight value corresponding to the solidity value of the relation edge, the relation edge merging value can be determined based on the shortest path value, the relation edge solidity value, the first weight value, and the second weight value of the relation edges to each node. For example, the product of the shortest path value and the first weight value is recorded as the first relation edge value, and the product of the relation edge solidity value and the second weight value is recorded as the second relation edge value. Then, the sum of the first relation edge value and the second relation edge value corresponding to the relation edge of the same node is recorded as the relation edge merging value corresponding to the relation edge of that node.

[0097] Furthermore, the merged value of the relation edges corresponding to each node's relation edges can be determined using the following expression:

[0098] V ij =α*I ij +(1-α)*B ij

[0099] Among them, V ijThis refers to the combined value of the relationship edges corresponding to the node relationship edges connecting sample image node i and sample image node j; α is the normalization coefficient, which can be set according to the sample image nodes contained in the image connected subgraph (α is the second weight value mentioned above, and 1-α is the first weight value mentioned above); I ij That is, the robustness coefficient of the relational edge corresponding to the node relational edge connecting sample image node i and sample image node j; B ij This is the shortest path value corresponding to the node relationship edge connecting sample image node i and sample image node j.

[0100] S502: Determine whether the minimum value of the merged relation edge is greater than or equal to the preset clustering threshold.

[0101] Understandably, the preset clustering threshold can be set based on the sample image nodes contained in the image connected subgraph.

[0102] S503: If the minimum value of the merged relation edge is greater than or equal to the preset clustering threshold, then the image connected graph is recorded as the target connected graph.

[0103] Specifically, after determining the merging value of the relation edges corresponding to each node based on the shortest path value and the stability value of the relation edges, a preset clustering threshold is obtained. The minimum merging value of the relation edges is then compared with the preset clustering threshold. If the minimum merging value of the relation edges is greater than or equal to the preset clustering threshold, it indicates that the density between sample image nodes in the existing image connectivity graph is good and the stability of the node relation edges is high. Therefore, the image connectivity graph can be directly recorded as the target connectivity graph.

[0104] S504: If the minimum value of the merged relation edge is less than the preset clustering threshold, then the node relation edge corresponding to the maximum value of the merged relation edge is recorded as the relation edge to be deleted, and the relation edge to be deleted is deleted from the image connected graph. Continue to determine the value of the merged relation edge corresponding to the node relation edge in the image connected graph after deleting the relation edge to be deleted, until the minimum value of the merged relation edge is greater than or equal to the preset clustering threshold. Record the image connected graph corresponding to the minimum value of the merged relation edge being greater than or equal to the preset clustering threshold as the target connected graph.

[0105] Specifically, when the minimum value of the merged relation edge is less than the preset clustering threshold, the density of the image connectivity graph and the stability of the node relation edges can be improved, meaning that the image connectivity graph can continue to be clustered and segmented. In the above description, when determining the shortest path value corresponding to the node relationship edge, if the shortest path value of a node relationship edge is greater than the preset shortest path edge number threshold, the node relationship edge corresponding to the largest shortest path value should be deleted. Furthermore, when determining the relationship edge strength coefficient, it indicates that the larger the relationship edge strength coefficient, the more unstable the corresponding node relationship edge. Thus, when the minimum relationship edge merging value is less than the preset clustering threshold, the node relationship edge corresponding to the largest relationship edge merging value is recorded as the relationship edge to be deleted, and deleted from the image connected graph. Following the above steps, the relationship edge merging value corresponding to the node relationship edge in the image connected graph after deleting the relationship edge to be deleted is determined. That is, the image connected graph after deleting the relationship edge to be deleted is re-segmented, and the shortest path value, relationship edge strength value, and relationship edge merging value of each node relationship edge in the image connected graph after deleting the relationship edge to be deleted are determined, until the minimum relationship edge merging value is greater than or equal to the preset clustering threshold. The image connected graph with the minimum relationship edge merging value greater than or equal to the preset clustering threshold is recorded as the target connected graph.

[0106] For example, assuming the minimum merged value of relation edges in the image connected graph after deleting the relation edges to be deleted is greater than or equal to a preset clustering threshold, the image connected graph after deleting the relation edges to be deleted can be recorded as the target connected graph. If the minimum merged value of relation edges in the image connected graph after deleting the relation edges to be deleted is still less than the preset clustering threshold, then the node relation edges corresponding to the maximum merged value of relation edges (this node relation edge is not the relation edge to be deleted in the above steps) are deleted, and the merged values ​​of relation edges of all node relation edges in the new image connected graph after deletion are recalculated. If the minimum merged value of relation edges in the new image connected graph is greater than or equal to the preset clustering threshold, then the new image connected graph is recorded as the target connected graph.

[0107] In this embodiment, the merging value of the relationship edges with each node is determined based on the shortest path value and the robustness value of the relationship edges. A preset clustering threshold is introduced, and when the minimum merging value is greater than the preset clustering threshold, the target connected graph corresponding to the sample image cluster is determined from the image connected graph. In this way, the image connected graph can be clustered and segmented by combining the shortest path value and the robustness value of the relationship edges, which improves the accuracy and precision of clustering and segmentation.

[0108] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0109] In one embodiment, an image clustering apparatus is provided, which corresponds one-to-one with the image clustering methods described in the above embodiments. For example... Figure 6 As shown, the image clustering device includes a cluster acquisition module 10, an image clustering module 20, a connected graph segmentation module 30, a relation edge analysis module 40, and a connected graph segmentation module 50. The functional modules are described in detail below: The cluster acquisition module 10 is used to acquire sample image clusters; the sample image clusters include multiple sample image nodes; the image clustering module 20 is used to perform image clustering on each of the sample image nodes to obtain an image connected graph; the connected graph segmentation module 30 is used to segment the image connected graph to obtain at least one image connected subgraph; each image connected subgraph includes at least one node relation edge connecting two of the sample image nodes; the relation edge analysis module 40 is used to determine the shortest path value and the relation edge strength value corresponding one-to-one with all node relation edges in each of the image connected subgraphs; the connected graph segmentation module 50 is used to determine the target connected graph corresponding to the sample image cluster from the image connected graph based on the shortest path value and relation edge strength value corresponding to each node relation edge.

[0110] Preferably, the image clustering module 20 includes: a first image vertex selection unit, used to select a sample image node from all sample image nodes as an image vertex; an image similarity determination unit, used to determine the image similarity between the image vertex and all unselected sample image nodes respectively; a first node relationship edge generation unit, used to select a preset number of sample image nodes with the highest image similarity from all unselected sample image nodes according to the image similarity, as the image connection nodes corresponding to the image vertex, and to connect the image vertex with the image connection nodes corresponding to the image vertex with directed edges to obtain node relationship edges; a second image vertex selection unit, used to reselect a sample image node from all sample image nodes as a new image vertex, the new sample image node being a sample image node that has not been selected from all sample image nodes; a second node relationship edge generation unit, used to continue to determine the node relationship edges corresponding to the new image vertex until all sample image nodes have been selected; and an image connected graph generation unit, used to generate the image connected graph based on all obtained node relationship edges.

[0111] Preferably, the relation edge analysis module 40 includes: a shortest path determination unit, used to determine all connected shortest paths in the image connected subgraph; a connected shortest path refers to the shortest path between any two sample image nodes in the image connected subgraph; a connected shortest path contains at least one node relation edge; a relation edge path value determination unit, used to record the total number of times the node relation edge appears in all the connected shortest paths as the relation edge path value corresponding to the node relation edge; a value comparison unit, used to determine whether there is at least one node relation edge whose relation edge path value is greater than a preset shortest path edge number threshold; and a first shortest path value determination unit, used to determine if there is at least one relation edge path value greater than the preset shortest path edge number threshold. For node relationship edges with a path edge count threshold, the path value of the relationship edge corresponding to each of the node relationship edges is recorded as the corresponding shortest path value; the second shortest path value determination unit is used to, if there is at least one node relationship edge with a path value greater than the preset shortest path edge count threshold, record the node relationship edge corresponding to the largest path value as the largest relationship edge, delete the largest relationship edge from the connected subgraph of the image, and continue to determine the path values ​​of the relationship edges corresponding to the node relationship edges of the connected subgraph of the image after deleting the largest relationship edge, until there is no node relationship edge with a path value greater than the preset shortest path edge count threshold, and record the path values ​​of the relationship edges corresponding to each of the node relationship edges as the corresponding shortest path values.

[0112] Preferably, the relation edge analysis module 40 further includes: a node recording unit, used to record two sample image nodes connected by the node relation edge as a first image node and a second image node, respectively; a node firmness coefficient determination unit, used to determine a first node firmness coefficient of the first image node in its respective image connected subgraph, and a second node firmness coefficient of the second image node in its respective image connected subgraph; and a relation edge firmness coefficient determination unit, used to determine the relation edge firmness coefficient corresponding to the node relation edge based on the first node firmness coefficient and the second node firmness coefficient.

[0113] Preferably, the node stability coefficient determination unit includes: a first node in-degree acquisition subunit, used to acquire the node out-degree and node in-degree of the first image node in its respective image connected subgraph; a first node stability coefficient determination subunit, used to record the sum of the node out-degree and the node in-degree as an in-degree sum, and determine the first node stability coefficient of the first image node in its respective image connected subgraph based on the node in-degree and the in-degree sum; a second node in-degree acquisition subunit, used to acquire the second node out-degree and the second node in-degree of the second image node in its respective image connected subgraph; a second node stability coefficient determination subunit, used to record the sum of the second node out-degree and the second node in-degree as a second in-degree sum, and determine the second node stability coefficient of the second image node in its respective image connected subgraph based on the second node in-degree and the second in-degree sum.

[0114] Preferably, the connected graph segmentation module 50 includes: a relation edge merging value determination unit, used to determine the relation edge merging value corresponding to each of the node relation edges based on the shortest path value and the relation edge firmness value; a value comparison unit, used to determine whether the smallest relation edge merging value is greater than or equal to a preset clustering threshold; a first target connected graph determination unit, used to record the image connected graph as the target connected graph if the smallest relation edge merging value is greater than or equal to the preset clustering threshold; and a second target connected graph determination unit, used to record the node relation edge corresponding to the largest relation edge merging value as a relation edge to be deleted if the smallest relation edge merging value is less than the preset clustering threshold, and delete the relation edge to be deleted from the image connected graph, and continue to determine the relation edge merging value corresponding to the node relation edge in the image connected graph after deleting the relation edge to be deleted, until the smallest relation edge merging value is greater than or equal to the preset clustering threshold, and record the image connected graph with the smallest relation edge merging value greater than or equal to the preset clustering threshold as the target connected graph.

[0115] In one embodiment, the relation edge merging value determination unit includes: a weight acquisition subunit, configured to acquire a first weight value corresponding to the shortest path value and a second weight value corresponding to the relation edge solidity value; a relation edge value determination subunit, configured to record the product of the shortest path value and the first weight value as a first relation edge value, and record the product of the relation edge solidity value and the second weight value as a second relation edge value; and a relation edge merging value determination subunit, configured to record the sum of the first relation edge value and the second relation edge value corresponding to the relation edge of the same node as the relation edge merging value corresponding to the relation edge of that node.

[0116] Specific limitations regarding the image clustering device can be found in the limitations of the image clustering method described above, and will not be repeated here. Each module in the aforementioned image clustering device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0117] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data used by the image clustering method in the above embodiments. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements an image clustering method.

[0118] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the image clustering method described above.

[0119] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the image clustering method described above.

[0120] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0121] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0122] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An image clustering method, characterized in that, include: Obtain a sample image cluster; the sample image cluster includes multiple sample image nodes; Image clustering is performed on each of the sample image nodes to obtain an image connectivity graph; The image connected graph is segmented to obtain at least one image connected subgraph; each image connected subgraph includes at least one node relationship edge connecting two sample image nodes; Determine the shortest path values ​​and the robustness values ​​of the relationship edges that correspond one-to-one with all node relationship edges in each of the image connected subgraphs; Based on the shortest path value and the firmness value of the relationship edge corresponding to each node, a target connected graph corresponding to the sample image cluster is determined from the image connected graph. The determination of the shortest path value corresponding one-to-one with all node relationship edges in each of the connected subgraphs of the image includes: Determine all connected shortest paths in the connected subgraph of the image; a connected shortest path is the shortest path between any two sample image nodes in the connected subgraph of the image; a connected shortest path contains at least one node relationship edge; Record the total number of times the node relationship edge appears in all the connected shortest paths as the relationship edge path value corresponding to that node relationship edge; Determine whether there exists at least one node relationship edge whose path value is greater than a preset shortest path edge number threshold. If there is no node relationship edge whose path value is greater than the preset shortest path edge number threshold, then the path value of the relationship edge corresponding to each node relationship edge is recorded as the corresponding shortest path value. If there is at least one node relation edge whose path value is greater than the preset shortest path edge number threshold, then the node relation edge corresponding to the largest relation edge path value is recorded as the largest relation edge, and the largest relation edge is deleted from the connected subgraph of the image. Continue to determine the relation edge path value corresponding to the node relation edge of the connected subgraph of the image after deleting the largest relation edge, until there is no node relation edge whose path value is greater than the preset shortest path edge number threshold. Then, the relation edge path value corresponding to each node relation edge is recorded as the corresponding shortest path value. The determination of the robust values ​​of the relation edges that correspond one-to-one with all node relation edges in each of the image connected subgraphs includes: Record the two sample image nodes connected by the node relationship edge as the first image node and the second image node, respectively. Determine the first node stability coefficient of the first image node in its respective image connected subgraph, and the second node stability coefficient of the second image node in its respective image connected subgraph; the first node stability coefficient is determined based on the node out-degree and node in-degree of the first image node in its respective image connected subgraph, and the second node stability coefficient is determined based on the node out-degree and node in-degree of the second image node in its respective image connected subgraph. Based on the first node's stability coefficient and the second node's stability coefficient, determine the stability coefficient of the relation edge corresponding to the node's relation edge.

2. The image clustering method as described in claim 1, characterized in that, The step of performing image clustering on each of the sample image nodes to obtain an image connectivity graph includes: Select one sample image node from all sample image nodes as the image vertex; Determine the image similarity between the image vertex and all unselected sample image nodes respectively; Based on the image similarity, among all unselected sample image nodes, the preset number of sample image nodes with the highest image similarity are selected as the image connection nodes corresponding to the image vertex, and the image vertex is connected to the image connection node corresponding to the image vertex by a directed edge to obtain the node relationship edge; A new sample image node is selected from all sample image nodes as a new image vertex. The new sample image node is a sample image node that has not been selected from all sample image nodes. Continue to determine the node relationship edges corresponding to the new image vertices until all sample image nodes have been selected; Based on all the obtained node relationship edges, the image connectivity graph is generated.

3. The image clustering method as described in claim 1, characterized in that, Determining the first node robustness coefficient of the first image node in its respective image connected subgraph includes: Obtain the out-degree and in-degree of the first image node in its respective image connected subgraph; The sum of the out-degree and in-degree of the first node is recorded as the first sum of out-degree and in-degree. Based on the in-degree of the first node and the first sum of out-degree and in-degree, the first node stability coefficient of the first image node in its respective image connected subgraph is determined. The determination of the second node robustness coefficient of the second image node in its respective image connected subgraph includes: Obtain the out-degree and in-degree of the second image node in its associated connected subgraph; The sum of the out-degree and in-degree of the second node is recorded as the second sum of in-degree and out-degree. Based on the in-degree of the second node and the second sum of in-degree and out-degree, the second node stability coefficient of the second image node in its respective image connected subgraph is determined.

4. The image clustering method as described in claim 1, characterized in that, The step of determining the target connected graph corresponding to the sample image cluster from the image connected graph based on the shortest path value and the firmness value of the connection edge corresponding to each of the node relationships includes: Based on the shortest path value and the firmness value of the relationship edge corresponding to each of the nodes, the relationship edge merging value corresponding to each of the nodes is determined; Determine whether the minimum value of the merged relation edge is greater than or equal to the preset clustering threshold; If the minimum value of the merged relation edge is greater than or equal to the preset clustering threshold, then the image connected graph is recorded as the target connected graph; If the minimum merged relation edge value is less than the preset clustering threshold, then the node relation edge corresponding to the maximum merged relation edge value is recorded as the relation edge to be deleted, and the relation edge to be deleted is deleted from the image connected graph. The merged relation edge value corresponding to the node relation edge in the image connected graph after deleting the relation edge to be deleted is determined until the minimum merged relation edge value is greater than or equal to the preset clustering threshold. The image connected graph when the minimum merged relation edge value is greater than or equal to the preset clustering threshold is recorded as the target connected graph.

5. The image clustering method as described in claim 4, characterized in that, The step of determining the relationship edge merging value corresponding to each of the node's relationship edges based on the shortest path value and the relationship edge strength value includes: Obtain a first weight value corresponding to the shortest path value, and a second weight value corresponding to the robustness value of the relation edge; The product of the shortest path value and the first weight value is recorded as the first relation edge value, and the product of the relation edge strength value and the second weight value is recorded as the second relation edge value. The sum of the values ​​of the first and second relation edges corresponding to the same node is recorded as the merged relation edge value corresponding to that node.

6. An image clustering device, characterized in that, include: A cluster acquisition module is used to acquire a sample image cluster; the sample image cluster includes multiple sample image nodes. The image clustering module is used to perform image clustering on each of the sample image nodes to obtain an image connectivity graph; A connected graph segmentation module is used to segment the image connected graph to obtain at least one image connected subgraph; one of the image connected subgraphs includes at least one node relationship edge connecting two of the sample image nodes; The relation edge analysis module is used to determine the shortest path value and the robustness value of the relation edge corresponding one-to-one with all node relation edges in each of the image connected subgraphs. The connected graph segmentation module is used to determine the target connected graph corresponding to the sample image cluster from the image connected graph based on the shortest path value and the firmness value of the relation edges corresponding to each node. The determination of the shortest path value corresponding one-to-one with all node relationship edges in each of the connected subgraphs of the image includes: Determine all connected shortest paths in the connected subgraph of the image; a connected shortest path is the shortest path between any two sample image nodes in the connected subgraph of the image; a connected shortest path contains at least one node relationship edge; Record the total number of times the node relationship edge appears in all the connected shortest paths as the relationship edge path value corresponding to that node relationship edge; Determine whether there exists at least one node relationship edge whose path value is greater than a preset shortest path edge number threshold. If there is no node relationship edge whose path value is greater than the preset shortest path edge number threshold, then the path value of the relationship edge corresponding to each node relationship edge is recorded as the corresponding shortest path value. If there is at least one node relation edge whose path value is greater than the preset shortest path edge number threshold, then the node relation edge corresponding to the largest relation edge path value is recorded as the largest relation edge, and the largest relation edge is deleted from the connected subgraph of the image. Continue to determine the relation edge path value corresponding to the node relation edge of the connected subgraph of the image after deleting the largest relation edge, until there is no node relation edge whose path value is greater than the preset shortest path edge number threshold. Then, the relation edge path value corresponding to each node relation edge is recorded as the corresponding shortest path value. The determination of the robust values ​​of the relation edges that correspond one-to-one with all node relation edges in each of the image connected subgraphs includes: Record the two sample image nodes connected by the node relationship edge as the first image node and the second image node, respectively. Determine the first node stability coefficient of the first image node in its respective image connected subgraph, and the second node stability coefficient of the second image node in its respective image connected subgraph; the first node stability coefficient is determined based on the node out-degree and node in-degree of the first image node in its respective image connected subgraph, and the second node stability coefficient is determined based on the node out-degree and node in-degree of the second image node in its respective image connected subgraph. Based on the first node's stability coefficient and the second node's stability coefficient, determine the stability coefficient of the relation edge corresponding to the node's relation edge.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the image clustering method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the image clustering method as described in any one of claims 1 to 5.

Images