JP2023502863A - Image incremental clustering method and apparatus, electronic device, storage medium and program product - Google Patents

Abstract

The present disclosure provides a method and apparatus for incremental clustering of images, an electronic device, a storage medium and a program product, the method obtains a first cluster of a first image data set, divides the first cluster into M dividing into one sub-cluster, obtaining a first cluster center corresponding to each first sub-cluster of the M first sub-clusters, wherein M is an integer of 1 or more, and a second image data set is obtained; and merging the second image data set and the first cluster using the first cluster center.

Description

(Cross reference to related applications)
This disclosure is proposed based on and claims priority from a Chinese patent application with filing number 202011185911.8 and filing date of October 30, 2020, and All are incorporated into this disclosure by reference.

TECHNICAL FIELD Embodiments of the present disclosure relate to the technical field of computer vision, and more particularly to methods and apparatus for incremental clustering of images, electronic devices, storage media, and program products.

Image processing technology has made great progress with the development of deep learning. Taking face recognition as an example, the recognition accuracy of face recognition models obtained by supervised learning has improved dramatically, but it has also exploded. How to classify accurately and quickly in the face of increasing amounts of unlabeled image data is still an issue worthy of discussion and research.

In view of the above problems, the present disclosure provides an image incremental clustering method and apparatus, an electronic device, a storage medium and a program product to solve the problem that the clustering effect is affected by cluster center drift in incremental clustering. Helpful.

To achieve the above objectives, a first aspect of embodiments of the present disclosure provides an incremental clustering method for images. The incremental clustering method for the image comprises:
obtaining a first cluster of a first image data set; dividing said first cluster into M first sub-clusters, corresponding to each first sub-cluster of said M first sub-clusters; obtaining a first cluster center, wherein M is an integer greater than or equal to 1; obtaining a second image data set; merging with one cluster.

According to a first aspect, in one possible embodiment said first cluster comprises a first cluster A, a first cluster B and a first cluster C, and said first cluster center is used to The step of merging the first cluster with
if the second image data set includes a plurality of image data, clustering the plurality of image data to obtain isolated image data and a second cluster; and clustering the isolated image data using the first cluster center. merging into said first cluster A and merging said second cluster with said first cluster B using said first cluster center; and only a single image data exists in said second image data set. if so, merging the single image data into the first cluster C using the first cluster center.

In this way, a plurality of image data in the second image data set are clustered, and the obtained isolated image data and the second cluster, the first cluster A, the first cluster B, and the first cluster included in the first cluster By merging C and C respectively, incorporation of single samples into clusters and merging between clusters can be realized.

According to the first aspect, in one possible embodiment, the first cluster has a corresponding second cluster center, and the first cluster center is used to divide the second image data set and the first cluster. Before merging, the incremental clustering method includes:
Further comprising determining K first clusters from the first cluster using the second cluster center.

According to a first aspect, in one possible embodiment, said second cluster has a corresponding third cluster center, and said second cluster center is used to derive K first clusters from said first cluster. The step of determining
obtaining a first similarity between the isolated image data and the center of the second cluster, sorting the first clusters in ascending order based on the first similarity to obtain a first cluster series, selecting K first clusters from the beginning in a cluster series; obtaining a second similarity between the third cluster center and the second cluster center; and based on the second similarity, the first sorting the clusters in ascending order to obtain a second cluster sequence, and selecting the top K first clusters in said second cluster sequence; or said single image data and said Obtaining a third similarity with the second cluster center, sorting the first cluster in ascending order based on the third similarity to obtain a third cluster sequence, K from the top in the third cluster sequence selecting first clusters.

Thus, using the calculated similarities between the center of the second cluster and the isolated image data, the center of the third cluster, and the single image data, the selection of the first cluster is performed by selecting the second image data It helps determine the first cluster that is closer to the cluster category of the image data in the set.

According to a first aspect, in one possible embodiment, the step of merging said isolated image data and said first cluster A using said first cluster center comprises:
obtaining a fourth similarity between the isolated image data and a first cluster center D, wherein the first cluster center D is the first being the first cluster center corresponding to a sub-cluster; and for each first cluster of the K first clusters, within each of the first clusters, the fourth similarity being a first threshold; and determining the first cluster among the K first clusters with the largest first quantity as the first cluster A. and merging the isolated image data and the first cluster A.

Thus, the first sub-cluster closer to the isolated image data is most present in the first cluster A, and merging the isolated image data into the first cluster A can make the clustering result more accurate. .

According to a first aspect, in one possible embodiment, the step of merging said second cluster and said first cluster B using said first cluster center comprises:
dividing the second cluster into N second sub-clusters and obtaining a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters, wherein N is being an integer greater than or equal to 1; and obtaining a fifth similarity between the fourth cluster center and the first cluster center E, wherein the first cluster center E is one of the K first clusters. and for each first cluster of the K first clusters, within each first cluster, determining a second quantity of said first cluster centers E for which said fifth similarity is greater than a second threshold; determining as the first cluster B; and merging the second cluster and the first cluster B;

Thus, the first cluster K has the highest quantity and we determine it as the first cluster B, i.e. the first sub-cluster closer to the second sub-cluster of the second cluster is the most abundant in the first cluster B and merging the second cluster into the first cluster B, the clustering result can be made more accurate.

According to the first aspect, in one possible embodiment, the step of merging said single image data and said first cluster C using said first cluster center comprises:
obtaining a sixth similarity measure between the single image data and a first cluster center F, wherein the first cluster center F is for each first cluster of the K first clusters; being the first cluster center corresponding to a first sub-cluster; and for each first cluster of the K first clusters, within each first cluster, the sixth similarity being the first determining a third quantity of the first cluster center F that is greater than a threshold of 3; and determining the first cluster among the K first clusters with the largest third quantity as the first cluster C. and merging the single image data and the first cluster C.

In this way, the first sub-cluster closer to the single image data exists most in the first cluster C, and by merging the single image data into the first cluster C, the clustering result is more accurate. can do.

According to the first aspect, in one possible embodiment, said M is less than or equal to a fourth threshold, and after merging said second image data set and said first cluster using said first cluster center, said The incremental clustering method is
dividing the merged first cluster into R third sub-clusters and obtaining a fifth cluster center of each third sub-cluster of said R third sub-clusters, wherein said R is 1 and if R is less than or equal to the fourth threshold, then retain the R third subclusters and use the fifth cluster center corresponding to the R third subclusters. updating the first cluster center; and obtaining a fourth quantity of image data in each third sub-cluster of the R third sub-clusters if the R is greater than the fourth threshold. sorting the R third sub-clusters according to the fourth quantity to obtain a fourth cluster sequence, and selecting the first P third sub-clusters in the fourth cluster sequence; and updating the first cluster center using the fifth cluster center corresponding to the P third subclusters, wherein the P is less than or equal to the fourth threshold. .

Thus, if there are many sub-clusters, leaving sub-clusters with more image data can limit the amount of sub-centers and remove the effect of outlier image data, thereby: It not only facilitates maintenance, but also enables high clustering effects to be obtained even in long-time large-scale incremental clustering scenes.

According to the first aspect, in one possible embodiment, said first cluster is obtained by clustering image data in said first image data set, said first cluster being divided into M first The step of dividing into sub-clusters is
obtaining a seventh degree of similarity between image data in the first cluster to obtain a similarity matrix; and dividing the first cluster into the M first sub-clusters based on the similarity matrix. and including.

Thus, a similarity matrix may be used to divide the first cluster into the M first sub-clusters.

According to the first aspect, in one possible embodiment the step of dividing said first cluster into said M first sub-clusters based on said similarity matrix comprises:
acquiring a connectivity graph configured with image data in the first cluster as vertices; searching from the similarity matrix to obtain the seventh degree of similarity between vertices of the connectivity graph; dividing a plurality of vertices whose similarity is greater than a fifth threshold as one first sub-cluster to obtain the M first sub-clusters.

In this way, a connectivity graph may be used to divide a plurality of vertices for which the seventh similarity is greater than the fifth threshold as one first sub-cluster.

A second aspect of embodiments of the present disclosure provides an apparatus for incremental clustering of images. The apparatus for incremental clustering of images comprises:
a first acquisition module configured to acquire a first cluster of a first image data set; dividing said first cluster into M first sub-clusters; a first segmentation module configured to obtain a first cluster center corresponding to each first subcluster, wherein M is an integer greater than or equal to 1; and obtaining a second image data set. and a merge module configured to merge the second image data set and the first cluster using the first cluster center.

A third aspect of embodiments of the present disclosure provides an electronic apparatus, including an input device and an output device, a processor suitable for implementing one or more instructions, and a processor loaded by the processor. a computer storage medium having stored therein one or more instructions suitable for performing the steps of any embodiment of the first aspect above.

A fourth aspect of embodiments of the present disclosure provides a computer storage medium, said computer storage medium, one or more suitable for being loaded by a processor to perform the steps of any of the embodiments of the first aspect above. instructions are stored.

A fifth aspect of embodiments of the present disclosure provides a computer program product, said computer program product being one or more suitable to be loaded by a processor to perform the steps of any of the embodiments of the first aspect above. including instructions for

As can be seen from the above, embodiments of the present disclosure obtain a first cluster of a first image data set, divide the first cluster into M first sub-clusters, and divide the M first sub-clusters into wherein M is an integer greater than or equal to 1, obtaining a second image data set, using the first cluster centers to obtain the second image Merge the dataset with the first cluster. Thus, the first cluster is divided into a plurality of first sub-clusters, and the merging of the first cluster and the second image data set is realized based on the first cluster center of the first sub-cluster, and the plurality of By maintaining the first cluster center (i.e. sub-center), with the increase of the image data, the cluster center (which is the cluster center of the first cluster, i.e. the main center) is newly added image data influence It solves the problem of drifting with , which helps to make the clustering result more accurate, and the clustering effect is improved. Also, in the clustering process, the second image data set does not need to calculate similarity to the entire first image data set, which helps reduce computational complexity.

1 is a schematic diagram of an application environment provided by an embodiment of the present disclosure; FIG. 4 is a flowchart of a method for incremental clustering of images provided by embodiments of the present disclosure; FIG. 4 is a schematic diagram of a connectivity graph of a first cluster provided by an embodiment of the present disclosure; FIG. 4 is a schematic diagram of dividing a first cluster into first sub-clusters provided by an embodiment of the present disclosure; FIG. 4 is a schematic diagram of a clustering result of a second image data set provided by an embodiment of the present disclosure; FIG. 4 is a schematic diagram of merging isolated image data into a first cluster provided by an embodiment of the present disclosure; FIG. 4 is a schematic diagram of merging a second cluster and a first cluster merging provided by an embodiment of the present disclosure; FIG. 4 is a flow chart of updating a first cluster center provided by an embodiment of the present disclosure; FIG. 4 is a flowchart of another incremental image clustering method provided by embodiments of the present disclosure; 1 is a block diagram of an incremental clustering device for images provided by an embodiment of the present disclosure; FIG. 1 is a configuration diagram of an electronic device provided by an embodiment of the present disclosure; FIG.

In order to make the solutions of the present disclosure better understood by those skilled in the art, the following clearly and completely describes the technical solutions in the embodiments of the present disclosure with reference to the drawings in the embodiments of the present disclosure, Of course, the described embodiments may not be all embodiments, but some of the embodiments of the present disclosure. All other embodiments obtained by persons skilled in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

The terms "including", "comprising" and any variations thereof in the specification, claims and drawings of this disclosure are intended to be non-exclusive. For example, a process, method, system, product or apparatus that includes a series of steps or units is not limited to the listed steps or units; or, in some embodiments of the present disclosure, further steps or units specific to these processes, methods or devices. Also, terms such as "first," "second," and "third," do not describe a particular order, but are intended to distinguish between different objects.

Incremental clustering is widely used in solving classification problems because images are often generated incrementally in real scenes, e.g. social media, security, etc. One cluster needs to be maintained, but different clusters have different sparseness, and with the progress of incremental clustering, the probability of cluster center drift increases and the clustering effect decreases.

An embodiment of the present disclosure provides an incremental clustering method for image data, the incremental clustering method can be implemented based on the application environment shown in FIG. 1, as shown in FIG. It includes a processing center 101 and an image acquisition device 102, the image processing center 101 including but not limited to a server 1011, terminals and databases. In some scenes, the image capture device 102 may be a camera or camera head placed in a scene such as a gated channel, shopping mall, residential area, etc. to capture images such as facial images, video surveillance images, etc. Well, the image processing center 101 may be a surveillance center, and the image processing center 101 may deploy a Video Cloud Node (VCN) 1012 to manage video surveillance, e.g. , and the images are clustered before being stored in database 1014 . In some scenes, the image collection device 102 may be a user terminal, the images collected by the image collection device 102 may be photos taken by the user, such as photos posted on social media by the user, The image processing center may be a social media processing background. The image acquisition device 102 may upload the acquired images to the image processing center 101, and the image processing center 101 may perform operations such as feature extraction, clustering classification, and face recognition. Since the generated and incremental clustering needs to maintain some clusters, as the image data increases and the incremental clustering progresses, there is a risk that the cluster centers of the originally maintained clusters will drift, which causes the clustering effect decreases gradually, so the server 1011 performs the incremental clustering method provided by the embodiments of the present disclosure to solve the problem that the clustering effect is affected by cluster center drift in incremental clustering. available to do. The server 1011 can be an independent physical server, a server cluster or a distributed system, and can also be cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, It may be a cloud server that provides basic cloud computing services such as middleware services, domain name services, security services, big data and artificial intelligence platforms.

In the following, the incremental clustering method for images provided by the embodiments of the present disclosure will be described in detail with reference to the associated drawings.

FIG. 2 is a flowchart of an incremental image clustering method provided by an embodiment of the present disclosure, which is applied to a server and includes steps S21 to S23, as shown in FIG.

At S21, the first cluster of the first image data set is obtained.

The first image data set refers to an image data set clustered as a plurality of clusters prior to the current batch of image data. ) is the current batch of data, the facial image data previously uploaded to the server is the first image data set. The first cluster is a cluster obtained by clustering the image data in the first image data set, the clustering algorithm employed may be a K-means clustering algorithm, and each cluster is It is understood to have a corresponding cluster center, ie a second cluster center.

At S22, divide the first cluster into M first sub-clusters, obtain a first cluster center corresponding to each first sub-cluster among the M first sub-clusters, wherein M is 1 is an integer equal to or greater than

FIG. 3A is a schematic diagram of a first cluster connectivity graph provided by an embodiment of the present disclosure, as shown in FIG. , a first cluster 301 is a cluster obtained by clustering the image data in the first image data set, and a second cluster center 302 is a corresponding cluster center of each cluster.

FIG. 3B is a schematic diagram of dividing a first cluster into first sub-clusters provided by an embodiment of the present disclosure, as shown in FIG. 3B, the first cluster is divided into first sub-clusters, A cluster 301, a second cluster center 302, a first sub-cluster 303 and a first cluster center 304 are included, the first sub-cluster 303 is a sub-cluster obtained by dividing the first cluster 301, and the first cluster Center 304 is the cluster center of each first sub-cluster.

The first sub-cluster is a sub-cluster obtained by dividing the first cluster. Obtain the similarity, obtain the similarity matrix, and then obtain the connection graph configured with the image data in the first cluster as the vertices, and as shown in FIG. For a vertex, search its similarity from the similarity matrix, and if the threshold used in clustering the first image data set is X, the fifth threshold, then the similarity is is larger than the X as one tighter first sub-cluster to obtain M first sub-clusters, and the first cluster shown in FIG. 3A is the connected graph The analysis divides into M first sub-clusters. After obtaining the M first sub-clusters, obtain the cluster center of each first sub-cluster of the M first sub-clusters, i.e. the first cluster center, whereby each first cluster has 1 It can be represented by one main cluster center and M sub-cluster centers. Representing the first cluster with more compact sub-clusters helps solve the problem of declining expressive power of a single main cluster center as newly added image data is incorporated.

At S23, a second image data set is obtained and the first cluster center is used to merge the second image data set and the first cluster.

FIG. 4A is a schematic diagram of the clustering result of the second image data set provided by the embodiment of the present disclosure, as shown in FIG. , a second cluster 402, an isolated image data 403 and a third cluster center 404, of which the second image data set 401 is a data set of images of the current batch uploaded by the image acquisition device, and the second cluster 402 is a cluster obtained by clustering the image data in the second image data set, isolated image data 403 is isolated image data that is not clustered, and third cluster center 404 is the center of each second cluster. is the cluster center with

FIG. 4B is a schematic diagram of merging isolated image data into a first cluster provided by an embodiment of the present disclosure, as shown in FIG. Cluster A 405 and isolated image data 403 are included, and the first cluster A 405 is the first cluster A determined from the first cluster.

FIG. 4C is a schematic diagram of merging the second cluster and the first cluster provided by an embodiment of the present disclosure, as shown in FIG. 4C, merging the second cluster and the first cluster; A cluster B 406 and a second cluster 407 are included, the first cluster B 406 and the second cluster 407 belonging to the same cluster category.

A second image dataset, ie, the dataset of the current batch of images uploaded by the image capture device, is derived from the images uploaded by the image capture device. The first cluster includes a first cluster A, a first cluster B and a first cluster C. When the second image data set includes a plurality of image data, the plurality of image data are clustered to obtain a clustering result. As shown in FIG. 4A, the clustering result includes unclustered isolated image data and a number of second clusters, each of the number of second clusters each having a corresponding cluster center , that is, has the third cluster center. For the isolated image data, determine the first cluster A from the first cluster, and merge it into the first cluster A using the first cluster center, i.e., as shown in FIG. 4B, the isolated image data is incorporated into the first cluster A, and the first cluster A and the isolated image data belong to the same cluster category. For each second cluster, determine the first cluster B from the first cluster and merge it into the first cluster B using the first cluster center, i.e. cluster and cluster , and the first cluster B and the second cluster belong to the same cluster category. Similar to the isolated image data, if there is only a single image data in the second image data set, i.e. if the newly added image data is single, a clustering operation is performed on the second image data set. , determine the first cluster C from the first cluster, merge it into the first cluster C using the first cluster center, and the first cluster C and the single image data are belong to the same cluster category.

In one possible embodiment, prior to merging the second image data set and the first clusters using the first cluster centers, the method for incremental clustering of images comprises:
Further comprising determining K first clusters from the first cluster using the second cluster center.

Before merging the second image data set and the first clusters, initially sort all the first clusters using the second cluster centers of the first clusters, and from all the first clusters, K first clusters and then select the above first cluster A and first cluster B or select the first cluster C from further K clusters. It should be explained that the K first clusters may be the top K after sorting all the first clusters using the second cluster center, for example sorting 100 first clusters and the K first clusters may be all first clusters after sorting, for example, after sorting 100 first clusters, still 100 select. Initial screening of the first clusters using the second cluster centers may result in the selection of the first clusters that are closer to the cluster category of the image data in the second image data set, e.g. It helps to determine one cluster C.

In one possible embodiment, determining K first clusters from said first cluster using said second cluster center comprises:
obtaining a first similarity between the isolated image data and the center of the second cluster, sorting the first clusters in ascending order based on the first similarity to obtain a first cluster sequence, and obtaining a first cluster sequence; selecting K first clusters from the beginning in the sequence; obtaining a second similarity between the third cluster center and the second cluster center; and based on the second similarity, the first cluster to obtain a second cluster series in descending order, and selecting K first clusters from the top in said second cluster series, or said single image data and said first Obtaining a third similarity with two cluster centers, sorting the first clusters in descending order based on the third similarity to obtain a third cluster sequence, and obtaining a third cluster sequence from the top K clusters in the third cluster sequence selecting the first cluster of .

When the second image data set is clustered to obtain isolated image data and a plurality of second clusters, a first similarity between each first cluster and a second cluster center is calculated for the isolated image data; For two clusters, calculate the second similarity between the corresponding third cluster center and the second cluster center of each first cluster, and based on the first similarity and the second similarity, respectively, all the first The clusters are sorted in descending order to obtain the corresponding first cluster series and second cluster series, and then the first K first clusters are selected from the first cluster series and the second cluster series, respectively. if only a single image data is included in the second image data set, calculating a third similarity between the single image data and the second cluster center of each first cluster; based on the third similarity, All the first clusters are sorted in descending order to obtain the corresponding third cluster series, and then the top K first clusters are selected from within the third cluster series.

In one possible embodiment, the step of merging the isolated image data and the first cluster A using the first cluster center comprises:
obtaining a fourth similarity between the isolated image data and a first cluster center D, wherein the first cluster center D is the first being the first cluster center corresponding to a sub-cluster; and for each first cluster of the K first clusters, within each of the first clusters, the fourth similarity being a first threshold; and determining the first cluster among the K first clusters with the largest first quantity as the first cluster A. and merging the isolated image data and the first cluster A.

For the merging of isolated image data, it is necessary to determine the first cluster A from the top K first clusters selected, and it should be explained that the top K first clusters are all may be the first cluster of First, the similarity between the isolated image data and the cluster center of each first sub-cluster of each first cluster out of the K first clusters (that is, the first cluster center D) is calculated and determined as the fourth similarity. and then analyzing the K first clusters, determining the quantity of first cluster centers D in each first cluster that satisfies that the fourth similarity is greater than the first threshold, and determining the first quantity and the first cluster with the largest first quantity is determined as the first cluster A. For example, among the K first clusters, the first cluster 1 has such a first cluster center D as 20 , the first cluster 2 has 18 such first cluster centers D, . . . , the first cluster K has 15 such first cluster centers D, and the first cluster 1 has a quantity is the most abundant, so it is determined as the first cluster A, i.e. the first sub-cluster that is closer to the isolated image data is the most present in the first cluster A, and the isolated image data is merged into the first cluster A. This makes the clustering results more accurate.

In one possible embodiment, the step of merging said second cluster and said first cluster B using said first cluster center comprises:
dividing the second cluster into N second sub-clusters and obtaining a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters, wherein N is being an integer greater than or equal to 1; and obtaining a fifth similarity between the fourth cluster center and the first cluster center E, wherein the first cluster center E is one of the K first clusters. and for each first cluster of the K first clusters, within each first cluster, determining a second quantity of said first cluster centers E for which said fifth similarity is greater than a second threshold; determining as the first cluster B; and merging the second cluster and the first cluster B;

For the merging of clusters and clusters, it is necessary to determine the first cluster B from the top K first clusters selected, and it should be explained that the top K first clusters are all may be the first cluster of First, divide each second cluster into N second sub-clusters in the manner of dividing the first cluster, calculate the cluster center of each second sub-cluster, i.e. the fourth cluster center, and then the fourth cluster calculating the similarity between the center and the cluster center of each first sub-cluster of each first cluster out of the K first clusters (i.e., the first cluster center E), and determining it as a fifth similarity; analyzing the K first clusters, determining the quantity of the first cluster center E in each first cluster that satisfies that the fifth similarity is greater than the second threshold, and determining it as the second quantity; The first cluster with the largest second quantity is determined as the first cluster B, for example, among the K first clusters, the first cluster 1 has 30 such first cluster centers E; One cluster 2 has 15 such first cluster centers E, ..., the first cluster K has 40 such first cluster centers E, and the first cluster K has the largest number, so Determine it as the first cluster B, i.e. the first sub-cluster closer to the second sub-cluster of the second cluster is most present in the first cluster B, merge the second cluster into the first cluster B This makes the clustering results more accurate.

In one possible embodiment, merging the single image data and the first cluster C using the first cluster center comprises:
obtaining a sixth similarity measure between the single image data and a first cluster center F, wherein the first cluster center F is for each first cluster of the K first clusters; being the first cluster center corresponding to a first sub-cluster; and for each first cluster of the K first clusters, within each first cluster, the sixth similarity being the first determining a third quantity of the first cluster center F that is greater than a threshold of 3; and determining the first cluster among the K first clusters with the largest third quantity as the first cluster C. and merging the single image data and the first cluster C.

For the merging of single image data, from the top K first clusters selected, the first cluster C needs to be determined. may be all the first clusters of First, the similarity between the single image data and the cluster center of each first sub-cluster of each first cluster out of the K first clusters (that is, the first cluster center F) is calculated, and a sixth similarity and then analyzing the K first clusters to determine the number of first cluster centers F in each first cluster that satisfy the sixth similarity greater than the third threshold; 3 quantity, and the first cluster with the largest third quantity is determined as the first cluster C, that is, the first sub-cluster that is closer to the single image data is present in the first cluster C the most. , by merging the single image data into the first cluster C, the clustering result can be more accurate.

In one possible embodiment, said M is less than or equal to a fourth threshold, and after merging said second image data set and said first cluster using said first cluster center, said The image incremental clustering method further includes the following steps.
In S51, divide the merged first cluster into R third sub-clusters, obtain a fifth cluster center of each third sub-cluster among the R third sub-clusters, wherein R is 1 or more; is an integer of
In S52, if the R is less than or equal to the fourth threshold, then the R third sub-clusters are retained, and the fifth cluster centers corresponding to the R third sub-clusters are used for the first cluster centers. to update.
At S53, if the R is greater than the fourth threshold, obtain a fourth quantity of image data in each third sub-cluster of the R third sub-clusters.
At S54, based on the fourth quantity, the R third sub-clusters are sorted in descending order to obtain a fourth cluster sequence, and P third sub-clusters are selected from the top in the fourth cluster sequence. and updating the first cluster center using the fifth cluster center corresponding to the P third sub-clusters, wherein P is less than or equal to the fourth threshold.

After merging the isolated image data and the second cluster into a first cluster, or after merging the single image data into a first cluster, the new image data is clustered in the original first cluster. , the subcenters of the original first cluster need to be updated. Specifically, in a method of splitting the first cluster, split the merged first cluster into R third sub-clusters, calculate the fifth cluster center of each third sub-cluster, and based on R Determine the quantity of 3 sub-clusters, and if the quantity of the 3rd sub-cluster is less than or equal to a fourth threshold, e.g., 20, then leave the R 3rd sub-clusters; Let the 5 cluster centers be the new sub-centers of the first cluster after merging, update the original first cluster center, and the first cluster after merging will be the second cluster center and R fifth cluster centers. expressed.

Also, if the quantity of the third sub-cluster is greater than the fourth threshold, sort the R third sub-clusters in descending order based on the quantity of image data in each third sub-cluster (i.e., the fourth quantity). Obtain a fourth cluster sequence, select and retain the top P third subclusters, for example, keep only the top 20 third subclusters, discard the remaining third subclusters, and select the P as the new sub-center of the first cluster after merging, update the original first cluster center, and the first cluster after merging is the second cluster center and P are represented by the fifth cluster centers. Each time a cluster is split into sub-clusters, only a preset number of sub-clusters are left, so both M and N are below the fourth threshold, thus the more sub-clusters, the more By leaving sub-clusters with image data, we can limit the amount of sub-centers and remove the effect of outlier image data, which not only makes maintenance easier, but also makes it easier to maintain large volumes over time. It is understood that it allows a high clustering effect to be obtained even in the scene of scale increment clustering.

As can be seen from the above, embodiments of the present disclosure obtain a first cluster of a first image data set, divide the first cluster into M first sub-clusters, and divide the M first sub-clusters into wherein M is an integer greater than or equal to 1, obtaining a second image data set, using the first cluster centers to obtain the second image Merge the dataset with the first cluster. Thus, the first cluster is divided into a plurality of first sub-clusters, and the merging of the first cluster and the second image data set is realized based on the first cluster center of the first sub-cluster, and the plurality of By maintaining the first cluster center (i.e. sub-center), with the increase of the image data, the cluster center (which is the cluster center of the first cluster, i.e. the main center) is newly added image data influence It solves the problem of drifting with , which helps to make the clustering result more accurate, and the clustering effect is improved. Also, in the clustering process, the second image data set does not need to calculate similarity to the entire first image data set, which helps reduce computational complexity.

Please refer to FIG. 6, which is a flowchart of another incremental clustering method for images provided by an embodiment of the present disclosure, including steps S61 to S66, as shown in FIG.
At S61, the first cluster of the first image data set is obtained.
At S62, divide the first cluster into M first sub-clusters, obtain a first cluster center corresponding to each first sub-cluster of the M first sub-clusters, where M is 1 is an integer equal to or greater than
At S63, a second image data set is obtained.
In S64, if the second image data set contains a plurality of image data, the plurality of image data are clustered to obtain isolated image data and a second cluster.
At S65, the isolated image data is merged into the first cluster A using the first cluster center, and the second cluster and the first cluster B are merged using the first cluster center.
At S66, if there is only a single image data in the second image data set, the single image data is merged into a first cluster C using the first cluster center.

The embodiments of steps S61 to S66 above have been described in the embodiments shown in FIGS. 2 to 5, and can achieve the same or similar beneficial effects, so detailed descriptions are omitted here.

Face recognition technology is constantly developing with breakthroughs in deep learning research, and face recognition models obtained by supervised learning have made breakthroughs. Accurate and rapid classification is a big issue with economic and social value.

In real scenes, such as social media, security, etc., the amount of image data is often large, and the data is generated incrementally every day, so incremental clustering methods are of greater practical use. have value. Incremental clustering schemes need to maintain several clusters in the clustering process, and conventional clustering algorithms represent a cluster by a single cluster center, e.g. Obtaining cluster centers, but different clusters have different sparseness, methods that simply use averaged single cluster centers tend to lose the rich sample information inside the clusters, and as the incremental clustering process progresses, , the clustering effect gradually becomes affected.

In the actual application process of face clustering, the facial features of different people have different distributions in the feature space data, some clusters have relatively compact samples inside, and some clusters have relatively sparse samples inside. . When a cluster is represented by a single center, the internal information of such a cluster can be lost, and as incremental clustering progresses, the influence of existing samples tapers off, and as new samples are added, the cluster center becomes The risk of drifting increases.

An image incremental clustering method provided by an embodiment of the present disclosure includes the following steps.

At S67, the similarity between cluster samples is calculated and a cluster is divided into several tighter sub-clusters.

Calculating the similarity between cluster samples yields a similarity matrix

を得ることができ、クラスタリングに使用される閾値が

and the threshold used for clustering is

であると仮定すると、より高い閾値

, the higher threshold

を設定する必要があり、つまり、

must be set, i.e.

を満たすことで、１つのクラスタをいくつかのより緊密なサブクラスタに分割する。

satisfies a cluster into several tighter sub-clusters.

The multicenter of the cluster may be obtained by analyzing the cluster in a manner based on connectivity graph analysis. By computing the cluster similarity matrix and using a similarity threshold higher than the one used for clustering, one cluster can be split into several tighter sub-clusters, thus: Multiple sub-cluster centers can be obtained, and the center of the cluster, which is the main center, is added to form a multi-center representation of the cluster.

Obtaining multiple sub-centers by multi-center design analysis of clustering based on connectivity graph analysis includes the following. First, for each cluster, by setting a higher threshold (which should be higher than the clustering threshold), we split the cluster into several more compact connected subgraphs, and sub-center , thereby obtaining multiple sub-centers, and the main center is still obtained by performing the conventional averaging method over the entire cluster.

At S68, in the process of incremental clustering, each time a new batch of data is added, the new data is clustered to generate a number of clusters and unclustered isolated samples.

At S69, some generated clusters and unclustered isolated samples are clustered and merged with the existing clustering results obtained at step S67.

The multi-center incremental clustering method based on a single main center and multiple sub-centers obtains the main center and multiple sub-centers, and then, in the process of incremental clustering, first obtains the main center and the new added data Use TopK to search and coarsely filter, and then determine whether to incorporate new samples or other clusters based on multiple subcenters.

This clustering-merging process involves merging between clusters and incorporating single isolated samples into clusters. For the incorporation of isolated sample points, based on the multi-center design, we first set a low threshold and search TopK at the main center, then the sub-centers and sample points are clustered with the clustering threshold

を満たすか否かを判断する。この場合、要件を満たす孤立サンプル点及びクラスタは複数存在し得、要件を満たすサブ中心数が最も多いクラスタを目標クラスタとする。クラスタ間のマージについては、同様に低い閾値でＴｏｐＫを選別してサーチし、続いて、クラスタ間に閾値要件を満たすサブ中心があるか否かを判断し、要件を満たすクラスタが複数存在する場合、閾値要件を満たすサブ中心数が最も多いクラスタを目標クラスタとする。

to determine whether or not it satisfies In this case, a plurality of isolated sample points and clusters satisfying the requirements may exist, and the cluster with the largest number of subcenters satisfying the requirements is set as the target cluster. For merging between clusters, we similarly filter and search TopK with a low threshold, and then determine whether there are subcenters between clusters that satisfy the threshold requirement, and if there are multiple clusters that satisfy the requirement, , the cluster with the largest number of subcenters satisfying the threshold requirement is taken as the target cluster.

Using a multicenter-based incremental clustering architecture, we synthetically use a single main center and multiple subcenters in the multicenter mechanism, and use the main centers to compute similarities when searching in TopK neighborhoods. , then compute the similarity using multiple sub-centers and a single sample or cluster awaiting clustering, thereby determining whether the incorporation of a single sample or the merging of clusters is complete. The architecture can comprehensively exploit the advantages of the multi-center representation scheme and improve the clustering effect without significantly increasing the computational complexity.

Upon merging clusters or adding new samples, the sub-centers need to be updated, and for simplicity of computation, may be modeled as sub-center clustering, which allows the merging and updating of sub-centers. come true. Also, to prevent too much data for a sub-center, each sub-center may be sorted in ascending order based on the number of sample points represented, e.g., select the 20 largest sub-centers You may make it

Use cluster multi-centric incremental update strategy. In an actual scene, as the amount of data continues to increase, merging and updating sub-centers and limiting the number of sub-centers can prevent excessive computation and storage burdens caused by the continuous increase in the number of sub-centers. , the influence of outlier interference points can be reduced.

Embodiments of the present disclosure fully take into account the complexities of face clustering based on large-scale data,
First, a multi-center construction scheme for face clusters is provided, in which a description of a face cluster with a single main center and multiple sub-centers can be obtained. The problem is that the cluster description maintains one cluster center and some compact sub-cluster information inside the cluster is omitted, and as the data continues to grow, a single cluster center is maintained. This solves the problem that the cluster center is continuously influenced by new samples, there is a risk of the center drifting, and the influence of the existing samples inside the cluster gradually weakens, reducing the expressive power of the center. And with a single cluster center, in the process of incremental clustering, the sample information inside the cluster is lost, and in the process of incremental clustering, we usually maintain a single cluster center for each cluster, and the data Merge and update clusters by calculating the similarity between cluster centers and new samples or clusters as we continue to add, and cluster centers are also continuously updated, and as we continue to add data, A single multicenter also solves the problem that the rich sample information inside the cluster is lost and is also prone to drift, thus affecting the clustering effect over time.

Next, a multicenter-based incremental clustering architecture is provided, which can strike a good balance between computational complexity and clustering accuracy in performing incremental clustering in terms of multicenter, and for a single sample Embedding into clusters and merging between clusters can be realized, solving the problem that the multi-center setting in the prior art greatly affects the computational speed and storage of clustering in large-scale data scenes.

Finally, a multi-center incremental update scheme is provided, which can achieve high clustering effect even in long-time large-scale incremental clustering scenes by merging and updating between sub-centers and limiting the number of sub-centers make it possible. The method can limit the increase in the number of multicenters and remove the influence of outliers. Preserving the center solves the problem of doubling the memory load when clustering, and the problem of an extra doubling of computation when searching in TopK neighborhoods.

According to the description of the method embodiment shown in FIG. 2 or FIG. 6, the embodiment of the present disclosure further provides an apparatus for incremental clustering of images, see FIG. 7, which is provided by the embodiment of the present disclosure. Fig. 7 is a block diagram of an incremental clustering device for images to be processed, and as shown in Fig. 7, the incremental clustering device for images is:
a first acquisition module 71 configured to acquire a first cluster of the first image data set;
a first division configured to divide the first cluster into M first sub-clusters and obtain a first cluster center corresponding to each first sub-cluster of the M first sub-clusters; a first division module 72, wherein M is an integer greater than or equal to 1;
a merge module 73 configured to obtain a second image data set and merge the second image data set and the first cluster using the first cluster center.

In one possible embodiment, said first cluster comprises a first cluster A, a first cluster B and a first cluster C, and said first cluster center is used to divide said second image data set and said first cluster. In terms of merging, the merging module 73 clusters the plurality of image data, if the second image data set includes a plurality of image data, to obtain isolated image data and a second cluster, the first cluster merging the isolated image data into the first cluster A using the center; merging the second cluster and the first cluster B using the first cluster center; into the second image data set It is configured to merge the single image data into the first cluster C using the first cluster center if only a single image data exists.

In one possible embodiment, the first cluster has a corresponding second cluster center, and before using the first cluster center to merge the second image data set and the first cluster, a merge module 73 is further configured to determine K first clusters from said first cluster using said second cluster center.

In one possible embodiment, the merge module, in that said second cluster has a corresponding third cluster center, and said second cluster center is used to determine K first clusters from said first cluster. 73 obtains a first similarity between the isolated image data and the center of the second cluster, sorts the first clusters in descending order based on the first similarity, and obtains a first cluster sequence; Select K first clusters from the top in the first cluster series, obtain a second similarity between the third cluster center and the second cluster center, and obtain the first similarity based on the second similarity sorting the clusters in ascending order to obtain a second cluster sequence, and selecting the top K first clusters in said second cluster sequence; or said single image data and said Obtaining a third similarity with the second cluster center, sorting the first cluster in ascending order based on the third similarity to obtain a third cluster sequence, K from the top in the third cluster sequence first clusters.

In one possible embodiment, merging module 73 merges the isolated image data with the first cluster center D in that it uses the first cluster center to merge the isolated image data with the first cluster center D. obtaining 4 similarities, wherein the first cluster center D is the first cluster center corresponding to each first sub-cluster of each first cluster of the K first clusters; determining, for each first cluster of the first clusters, a first quantity of said first cluster centers D within said each first cluster for which said fourth similarity is greater than a first threshold; The first cluster having the largest first quantity among the first clusters is determined as the first cluster A, and the isolated image data and the first cluster A are merged.

In one possible embodiment, in merging the second cluster and the first cluster B using the first cluster center, merge module 73 divides the second cluster into N second sub-clusters. dividing to obtain a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters, wherein N is an integer greater than or equal to 1, the fourth cluster center and the first cluster center; obtain a fifth similarity measure with E, wherein the first cluster center E is the first cluster center corresponding to each first sub-cluster of each first cluster of the K first clusters; for each first cluster of K first clusters, determine a second quantity of said first cluster centers E within said each first cluster for which said fifth similarity is greater than a second threshold; , determining the first cluster having the largest second quantity among the K first clusters as the first cluster B, and merging the second cluster and the first cluster B. be.

In one possible embodiment, in merging the single image data and the first cluster C using the first cluster center, the merge module 73 merges the single image data and the first cluster center obtaining a sixth similarity measure with F, wherein the first cluster center F is the first cluster center corresponding to each first sub-cluster of each first cluster of the K first clusters; For each first cluster of the K first clusters, determine a third quantity of the first cluster centers F within each first cluster for which the sixth similarity is greater than a third threshold. and determining the first cluster among the K first clusters having the largest third quantity as the first cluster C, and merging the single image data and the first cluster C. configured as

In one possible embodiment, the M is less than or equal to a fourth threshold, and the first splitting module 72 splits the merged first cluster into R third sub-clusters, and dividing the R third sub-clusters into Obtaining a fifth cluster center of each third sub-cluster of said updating the first cluster center with the fifth cluster center corresponding to R third sub-clusters, and if the R is greater than the fourth threshold, one of the R third sub-clusters; obtaining a fourth quantity of image data in each third sub-cluster, sorting the R third sub-clusters according to the fourth quantity in descending order to obtain a fourth cluster sequence; selecting P third subclusters from the top in the sequence, updating the first cluster center using the fifth cluster center corresponding to the P third subclusters, wherein the P is the fourth It is further configured to be less than or equal to the threshold.

In one possible embodiment, in terms of dividing the first cluster into M first sub-clusters, the first division module 72 obtains a seventh similarity measure between image data within the first clusters; It is configured to obtain a similarity matrix and divide the first cluster into the M first sub-clusters based on the similarity matrix.

In one possible embodiment, in dividing the first cluster into the M first sub-clusters based on the similarity matrix, the first division module 72 divides the image data in the first cluster into vertices and search from the similarity matrix to obtain the seventh similarity between vertices of the connected graph, and select the vertices where the seventh similarity is greater than a fifth threshold. partitioning as one first sub-cluster to obtain said M first sub-clusters.

According to one embodiment of the present disclosure, each or all of the units in the incremental image clustering apparatus shown in FIG. 7 may be configured integrally as one or more other units, or A certain (several) unit may be further divided into a plurality of functionally smaller units, both of which are capable of realizing the same operation, and to achieve the technical effects of the embodiments of the present disclosure. It does not affect. The above units are divided according to their logical functions. In practical application, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. good. In other embodiments of the present disclosure, the image-based incremental clustering device may include other units, and in actual application, these functions may be realized by cooperation of other units, and a plurality of It may be realized by cooperation of units.

According to another embodiment of the present disclosure, a general purpose computing device, such as a computer, including processing and memory elements such as a central processing unit (CPU), random access storage medium (RAM), read only storage medium (ROM), etc. Constructing an incremental image clustering device shown in FIG. 7 by executing a computer program (including program code) capable of executing the steps of the corresponding method shown in FIG. 2 or FIG. 6 above, The incremental clustering method for images of the embodiments of the present disclosure can be implemented. The computer program can be recorded in a computer-readable recording medium, loaded into the computing device via the computer-readable recording medium, and executed in the computing device, for example.

According to the above method embodiments and apparatus embodiments, embodiments of the present disclosure further provide an electronic device. Referring to FIG. 8, the electronic device includes at least a processor 81 , an input device 82 , an output device 83 and a computer storage medium 84 . The processor 81, input device 82, output device 83 and computer storage medium 84 within the electronic device may be connected via a bus or other manner.

Computer storage medium 84 may be stored in a memory of an electronic device, said computer storage medium 84 being configured to store a computer program including program instructions, said processor 81 being stored in said computer storage medium 84. configured to execute program instructions that A processor 81 (also called a CPU (Central Processing Unit)) is a computing core and control core of an electronic device, suitable for implementing one or more instructions, and executing one or more instructions. It is suitable for loading and executing to implement corresponding method flows or corresponding functions.

In one embodiment, processor 81 of an electronic device provided by embodiments of the present disclosure may be configured to perform the following incremental clustering process for a series of images.

obtaining a first cluster of the first image data set;
dividing the first cluster into M first sub-clusters, obtaining a first cluster center corresponding to each first sub-cluster among the M first sub-clusters, wherein M is an integer greater than or equal to 1; and obtaining a second image data set and merging the second image data set and the first cluster using the first cluster center.

In yet another embodiment, said first clusters include a first cluster A, a first cluster B and a first cluster C, and processor 81 uses said first cluster centers to determine said second image data set and said first cluster. performing the step of merging the image data with one cluster, and if the second image data set includes a plurality of image data, clustering the plurality of image data to obtain isolated image data and a second cluster; merging the isolated image data into the first cluster A using one cluster center and merging the second cluster and the first cluster B using the first cluster center; merging the single image data into the first cluster C using the first cluster center if there is only a single image data in the dataset.

In yet another embodiment, said first cluster has a corresponding second cluster center, and prior to merging said second image data set and said first cluster using said first cluster center, processor 81 is further configured to perform the step of determining K first clusters from said first cluster using said second cluster center.

In yet another embodiment, said second cluster has a corresponding third cluster center, and processor 81 determines K first clusters from said first cluster using said second cluster center. obtaining a first similarity between the isolated image data and the center of the second cluster, and sorting the first clusters in ascending order based on the first similarity to obtain a first cluster series; , selecting K first clusters from the top in the first cluster sequence; obtaining a second similarity between the third cluster center and the second cluster center; and based on the second similarity, , sorting the first clusters in ascending order to obtain a second cluster sequence, and selecting the top K first clusters in the second cluster sequence; or obtaining a third similarity between the image data and the center of the second cluster, sorting the first clusters in descending order based on the third similarity to obtain a third cluster sequence, and within the third cluster sequence selecting the K first clusters from the beginning of .

In yet another embodiment, processor 81 performs the step of merging said isolated image data with said first cluster A using said first cluster center, and merging said isolated image data with said first cluster center D. obtaining a fourth similarity measure, wherein the first cluster center D is the first cluster center corresponding to each first sub-cluster of each first cluster among the K first clusters; and for each first cluster of said K first clusters, the first of said first cluster centers D within said each first cluster for which said fourth similarity is greater than a first threshold. determining a quantity; determining a first cluster having the largest first quantity among the K first clusters as the first cluster A; and merging.

In yet another embodiment, processor 81 performs the step of merging said second cluster and said first cluster B using said first cluster center, dividing said second cluster into N second sub-clusters. and obtaining a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters, wherein N is an integer equal to or greater than 1; obtaining a fifth similarity between a cluster center and a first cluster center E, wherein the first cluster center E corresponds to each first sub-cluster of each first cluster among the K first clusters; and for each first cluster of the K first clusters, the fifth similarity within each first cluster is greater than a second threshold. determining a second quantity of the first cluster center E; determining a first cluster having the largest second quantity among the K first clusters as the first cluster B; merging a second cluster with said first cluster B;

In yet another embodiment, processor 81 performs the step of merging said single image data and said first cluster C using said first cluster center, said single image data and said first cluster obtaining a sixth similarity measure with a center F, wherein said first cluster center F corresponds to each first sub-cluster of each first cluster of said K first clusters; and for each first cluster of said K first clusters, said first cluster center within said each first cluster for which said sixth similarity is greater than a third threshold. determining a third quantity of F; determining a first cluster among the K first clusters having the largest third quantity as the first cluster C; and the single image data and said first cluster C.

In yet another embodiment, said M is less than or equal to a fourth threshold, and after merging said second image data set and said first cluster using said first cluster center, processor 81 calculates the merged first cluster center dividing a cluster into R third sub-clusters and obtaining a fifth cluster center of each third sub-cluster among the R third sub-clusters, wherein R is an integer greater than or equal to 1; a step, if said R is less than or equal to said fourth threshold, then said first cluster using said fifth cluster center corresponding to said R third sub-clusters, leaving said R third sub-clusters; updating the center; obtaining a fourth quantity of image data in each third sub-cluster of the R third sub-clusters if the R is greater than the fourth threshold; sorting the R third sub-clusters in descending order to obtain a fourth cluster sequence according to a fourth quantity; selecting the first P third sub-clusters in the fourth cluster sequence; updating the first cluster center with the fifth cluster center corresponding to the third sub-cluster, wherein the P is less than or equal to the fourth threshold be done.

In yet another embodiment, said first cluster is obtained by clustering image data in said first image data set, and processor 81 divides said first cluster into M first sub-clusters. performing a dividing step to obtain a seventh similarity between image data in the first cluster to obtain a similarity matrix; dividing the first cluster into the M th clusters based on the similarity matrix and dividing into one sub-cluster.

In yet another embodiment, processor 81 performs the step of dividing said first cluster into said M first sub-clusters based on said similarity matrix, with image data in said first cluster as vertices obtaining a constructed connectivity graph; searching from the similarity matrix to obtain the seventh similarity between vertices of the connectivity graph; and splitting the vertices as one first sub-cluster to obtain the M first sub-clusters.

Illustratively, the electronic device may be a computer, computer host, server, cloud server, server cluster, etc., although the electronic device includes a processor 81, an input device 82, an output device 83 and a computer storage medium 84. , without limitation, input device 82 may be a keyboard, touch screen, etc., and output device 83 may be a speaker, display, radio frequency transmitter, or the like. Those skilled in the art will appreciate that the schematic diagram may be an illustration of an electronic device, not a limitation of the electronic device, and may include more or fewer members than the drawings, or may combine any or different members. Good thing is understood.

It should be noted that the steps of the above-described incremental image clustering method are realized when the processor 81 of the electronic device executes the computer program, so that any of the above-described incremental image clustering method embodiments can be applied to the electronic device. applied and either can achieve the same or similar beneficial effects.

An embodiment of the present disclosure further provides a computer program product, which, when executed by a processor, implements the method of any one of the preceding embodiments. The computer program product can be implemented in hardware, software or a combination thereof. In some embodiments of the disclosure, the computer program product is embodied as a computer storage medium, and in other embodiments of the disclosure, the computer program product is software, such as a Software Development Kit (SDK). realized as a product.

Embodiments of the present disclosure further provide a computer storage medium (Memory), which is a storage device within an electronic device and is configured to store programs and data. It is understood that computer storage media herein may include internal storage media within the terminal as well as, of course, extended storage media supported by the terminal. A computer storage medium provides a storage space in which the operating system of the terminal is stored. Also stored in the memory space is one or more instructions suitable for being loaded and executed by processor 81, which instructions may be one or more computer programs (including program code). There may be. It should be noted that the computer storage medium herein may be high speed RAM memory or may be Non-Volatile Memory, such as at least one magnetic disk memory; , there may be at least one computer storage medium remotely located from the processor 81 . In one embodiment, one or more instructions stored in a computer storage medium may be loaded and executed by processor 81 to implement the corresponding steps of the incremental clustering of images method described above.

By way of example, a computer program on a computer storage medium comprises computer program code, which may be in source code form, object code form, an executable file, or some intermediate form, and the like. The computer-readable medium is any entity or device capable of carrying the computer program code, recording medium, USB flash disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random It may include access memory (RAM), electrical carrier signals, telecommunication signals, software distribution media, and the like.

It should be explained that the steps of the above incremental image clustering method are realized when the computer program on the computer storage medium is executed by the processor, so that all the embodiments of the above incremental image clustering method are applicable. applied to computer storage media, and either can achieve the same or similar beneficial effects.

Exemplary embodiments of the present disclosure have been described above in detail. Although the principles and embodiments of the present disclosure have been described herein using examples, the above examples are merely illustrative to facilitate understanding of the methods and core ideas of the present disclosure. A person skilled in the art can change both the embodiment and the scope of application based on the idea of the present invention, and therefore, the content of the present specification should not be understood as limiting the present disclosure.

In this embodiment, the first cluster is divided into a plurality of first sub-clusters, and the merging of the first cluster and the second image data set is realized based on the first cluster center of the first sub-cluster. By maintaining the first cluster center of , it solves the problem that as the image data increases, the cluster center drifts under the influence of the newly added image data, which makes the clustering result more accurate and improve the clustering effect.

As can be seen from the above, embodiments of the present disclosure obtain a first cluster of a first image data set, divide the first cluster into M first sub-clusters, and divide the M first sub-clusters into wherein M is an integer greater than or equal to 1, obtaining a second image data set, using the first cluster centers to obtain the second image Merge the dataset with the first cluster. Thus, the first cluster is divided into a plurality of first sub-clusters, and the merging of the first cluster and the second image data set is realized based on the first cluster center of the first sub-cluster, and the plurality of By maintaining the first cluster center (i.e. sub-center), with the increase of the image data, the cluster center (which is the cluster center of the first cluster, i.e. the main center) is newly added image data influence It solves the problem of drifting with , which helps to make the clustering result more accurate, and the clustering effect is improved. Also, in the clustering process, the second image data set does not need to calculate similarity to the entire first image data set, which helps reduce computational complexity.
For example, the present application provides the following items.
(Item 1)
obtaining a first cluster of the first image data set;
dividing the first cluster into M first sub-clusters and obtaining a first cluster center corresponding to each first sub-cluster of the M first sub-clusters, wherein M is a step that is an integer greater than or equal to 1;
obtaining a second image data set and merging the second image data set and the first cluster using the first cluster center.
(Item 2)
wherein said first cluster comprises a first cluster A, a first cluster B and a first cluster C, and merging said second image data set and said first cluster using said first cluster center comprises:
if the second image data set contains a plurality of image data, clustering the plurality of image data to obtain isolated image data and a second cluster;
merging the isolated image data into the first cluster A using the first cluster center and merging the second cluster and the first cluster B using the first cluster center;
merging the single image data into the first cluster C using the first cluster center if there is only a single image data in the second image data set.
A method for incremental clustering of images according to item 1.
(Item 3)
The first cluster has a corresponding second cluster center, and before using the first cluster center to merge the second image data set and the first cluster,
Further comprising determining K first clusters from the first cluster using the second cluster center.
A method for incremental clustering of images according to item 2.
(Item 4)
said second cluster having a corresponding third cluster center, and using said second cluster center to determine K first clusters from said first cluster;
obtaining a first similarity between the isolated image data and the second cluster center;
sorting the first clusters in descending order based on the first similarity to obtain a first cluster sequence, and selecting K first clusters from the top in the first cluster sequence;
obtaining a second similarity between the third cluster center and the second cluster center;
sorting the first clusters in descending order based on the second similarity to obtain a second cluster sequence, and selecting K first clusters from the top in the second cluster sequence. , or
obtaining a third similarity between the single image data and the second cluster center;
sorting the first clusters in descending order based on the third similarity to obtain a third cluster sequence, and selecting the top K first clusters in the third cluster sequence.
A method for incremental clustering of images according to item 3.
(Item 5)
merging the isolated image data and the first cluster A using the first cluster center,
obtaining a fourth similarity between the isolated image data and a first cluster center D, wherein the first cluster center D is the first being the first cluster center corresponding to a sub-cluster;
For each first cluster of the K first clusters, determine a first quantity of the first cluster centers D within each first cluster for which the fourth similarity is greater than a first threshold. and
determining a first cluster among the K first clusters having the largest first quantity as the first cluster A;
merging the isolated image data and the first cluster A.
A method for incremental clustering of images according to item 3.
(Item 6)
merging the second cluster and the first cluster B using the first cluster center,
dividing the second cluster into N second sub-clusters and obtaining a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters, wherein N is a step that is an integer greater than or equal to 1;
obtaining a fifth similarity between the fourth cluster center and the first cluster center E, wherein the first cluster center E is the first cluster center of each first cluster among the K first clusters; being the first cluster center corresponding to a sub-cluster;
For each first cluster of the K first clusters, determine a second quantity of the first cluster centers E within each first cluster for which the fifth similarity is greater than a second threshold. and
determining, as the first cluster B, a first cluster among the K first clusters that has the largest second quantity;
merging the second cluster and the first cluster B.
A method for incremental clustering of images according to item 3.
(Item 7)
merging the single image data and the first cluster C using the first cluster center;
obtaining a sixth similarity measure between the single image data and a first cluster center F, wherein the first cluster center F is for each first cluster of the K first clusters; being the first cluster center corresponding to a first sub-cluster;
For each first cluster of the K first clusters, determine a third quantity of the first cluster centers F within each first cluster for which the sixth similarity is greater than a third threshold. and
determining, as the first cluster C, the first cluster having the largest third quantity among the K first clusters;
merging the single image data and the first cluster C.
A method for incremental clustering of images according to item 3.
(Item 8)
wherein M is less than or equal to a fourth threshold, and after merging the second image data set and the first cluster using the first cluster center;
dividing the merged first cluster into R third sub-clusters and obtaining a fifth cluster center of each third sub-cluster of said R third sub-clusters, wherein said R is 1 a step that is an integer greater than or equal to
if the R is less than or equal to the fourth threshold, retain the R third sub-clusters and update the first cluster center using the fifth cluster center corresponding to the R third sub-clusters; a step;
obtaining a fourth quantity of image data in each third sub-cluster of the R third sub-clusters, if the R is greater than the fourth threshold;
sorting the R third sub-clusters in ascending order according to the fourth quantity to obtain a fourth cluster sequence, selecting P third sub-clusters from the top in the fourth cluster sequence; updating the first cluster center with the fifth cluster center corresponding to P third subclusters, wherein the P is less than or equal to the fourth threshold.
8. A method for incremental clustering of images according to any one of items 1-7.
(Item 9)
The first cluster is obtained by clustering the image data in the first image data set, and the step of dividing the first cluster into M first sub-clusters comprises:
obtaining a seventh similarity between image data in the first cluster to obtain a similarity matrix;
dividing the first cluster into the M first sub-clusters based on the similarity matrix.
8. A method for incremental clustering of images according to any one of items 1-7.
(Item 10)
dividing the first cluster into the M first sub-clusters based on the similarity matrix;
obtaining a connected graph configured with the image data in the first cluster as vertices;
searching from the similarity matrix to obtain the seventh similarity between vertices of the connectivity graph;
dividing a plurality of vertices whose seventh similarity is greater than a fifth threshold as one first sub-cluster to obtain the M first sub-clusters.
A method for incremental clustering of images according to item 9.
(Item 11)
a first acquisition module configured to acquire a first cluster of the first image data set;
a first division configured to divide the first cluster into M first sub-clusters and obtain a first cluster center corresponding to each first sub-cluster of the M first sub-clusters; a first divided module, wherein M is an integer of 1 or more;
a merge module configured to obtain a second image data set and merge the second image data set and the first cluster using the first cluster center.
(Item 12)
The first cluster includes a first cluster A, a first cluster B and a first cluster C, and the merge module comprises:
a clustering sub-module configured to cluster the plurality of image data to obtain isolated image data and a second cluster, if the second image data set includes a plurality of image data;
a first merging sub-module configured to merge the isolated image data into the first cluster A using the first cluster center;
a second merging sub-module configured to merge the second cluster and the first cluster B using the first cluster center;
a third merge configured to merge the single image data into the first cluster C using the first cluster center if only a single image data exists in the second image data set; contains submodules and
12. Apparatus for incremental clustering of images according to item 11.
(Item 13)
The first cluster has a corresponding second cluster center, the merge module comprising:
further comprising a first determining sub-module configured to determine K first clusters from the first cluster using the second cluster center;
13. Apparatus for incremental clustering of images according to item 12.
(Item 14)
said second cluster having a corresponding third cluster center, said first determining sub-module comprising:
a first obtaining unit configured to obtain a first similarity between the isolated image data and the second cluster center;
A first cluster configured to sort the first clusters in descending order based on the first similarity to obtain a first cluster sequence, and select K first clusters from the top in the first cluster sequence. 1 sorting unit;
a second obtaining unit configured to obtain a second similarity measure between the third cluster center and the second cluster center;
A second cluster configured to sort the first clusters in descending order based on the second similarity to obtain a second cluster sequence, and select K first clusters from the top in the second cluster sequence. 2 sorting units, or
a third obtaining unit configured to obtain a third similarity between the single image data and the second cluster center;
A third cluster configured to sort the first clusters in descending order based on the third similarity to obtain a third cluster sequence, and select K first clusters from the top in the third cluster sequence. 3 sorting units and
14. Apparatus for incremental clustering of images according to item 13.
(Item 15)
The first merging submodule includes:
a fourth obtaining unit configured to obtain a fourth similarity measure between the isolated image data and a first cluster center D, wherein the first cluster center D is one of the K first clusters; a fourth obtaining unit that is the first cluster center corresponding to each first sub-cluster of each first cluster;
For each first cluster of the K first clusters, determine a first quantity of the first cluster centers D within each first cluster for which the fourth similarity is greater than a first threshold. a first determining unit configured to
a second determining unit configured to determine the first cluster among the K first clusters with the largest first quantity as the first cluster A;
a first merging unit configured to merge the isolated image data and the first cluster A.
14. Apparatus for incremental clustering of images according to item 13.
(Item 16)
The second merging submodule includes:
A first division configured to divide the second cluster into N second sub-clusters and obtain a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters. a first divisional unit, wherein N is an integer of 1 or more;
a fifth obtaining unit configured to obtain a fifth similarity measure between the fourth cluster center and a first cluster center E, wherein the first cluster center E is one of K first clusters a fifth obtaining unit that is the first cluster center corresponding to each first sub-cluster of each first cluster;
For each first cluster of the K first clusters, determine a second quantity of the first cluster centers E within each first cluster for which the fifth similarity is greater than a second threshold. a third determining unit configured to
a fourth determining unit configured to determine the first cluster among the K first clusters with the largest second quantity as the first cluster B;
a second merging unit configured to merge the second cluster and the first cluster B.
14. Apparatus for incremental clustering of images according to item 13.
(Item 17)
The third merging submodule includes:
a sixth obtaining unit configured to obtain a sixth similarity measure between the single image data and a first cluster center F, wherein the first cluster center F is one of the K first clusters a sixth obtaining unit that is the first cluster center corresponding to each first sub-cluster of each first cluster thereof;
For each first cluster of the K first clusters, determine a third quantity of the first cluster centers F within each first cluster for which the sixth similarity is greater than a third threshold. a fifth determining unit configured to
a sixth determining unit configured to determine the first cluster among the K first clusters for which the third quantity is the largest as the first cluster C;
a third merging unit configured to merge the single image data and the first cluster C.
14. Apparatus for incremental clustering of images according to item 13.
(Item 18)
The M is equal to or less than the fourth threshold,
A second splitting module configured to split the merged first cluster into R third sub-clusters and obtain a fifth cluster center of each third sub-cluster of said R third sub-clusters. a second segmented module, wherein R is an integer greater than or equal to 1;
if the R is less than or equal to the fourth threshold, retain the R third sub-clusters and update the first cluster center using the fifth cluster center corresponding to the R third sub-clusters; a first update module configured to:
a second acquisition module configured to acquire a fourth quantity of image data in each third sub-cluster of the R third sub-clusters, if the R is greater than the fourth threshold;
sorting the R third sub-clusters in ascending order according to the fourth quantity to obtain a fourth cluster sequence, selecting P third sub-clusters from the top in the fourth cluster sequence; a second update module configured to update the first cluster center using the fifth cluster center corresponding to P third sub-clusters, wherein the P is less than or equal to the fourth threshold; 2 update modules, further including
Apparatus for incremental clustering of images according to any one of items 11 to 17.
(Item 19)
The first cluster is obtained by clustering the image data in the first image data set, and the first division module includes:
an acquisition sub-module configured to acquire a seventh degree of similarity between image data in the first cluster to obtain a similarity matrix;
a splitting sub-module configured to split the first cluster into the M first sub-clusters based on the similarity matrix.
Apparatus for incremental clustering of images according to any one of items 11 to 17.
(Item 20)
The divided sub-modules are
a seventh obtaining unit configured to obtain a connectivity graph configured with the image data in the first cluster as vertices;
a search unit configured to search from the similarity matrix to obtain the seventh similarity between vertices of the connectivity graph;
a second segmentation unit configured to segment a plurality of vertices with the seventh similarity greater than a fifth threshold as one first sub-cluster to obtain the M first sub-clusters.
20. Apparatus for incremental clustering of images according to item 19.
(Item 21)
including input devices and output devices;
a processor suitable for implementing one or more instructions;
and a computer storage medium loaded by said processor storing one or more instructions suitable for carrying out the method of any one of items 1 to 10.
(Item 22)
A computer storage medium loaded by a processor and storing one or more instructions suitable for carrying out the method of any one of items 1 to 10.
(Item 23)
A computer program product comprising one or more instructions suitable to be loaded by a processor to carry out the method of any one of items 1 to 10.

Claims (23)

obtaining a first cluster of the first image data set;
dividing the first cluster into M first sub-clusters and obtaining a first cluster center corresponding to each first sub-cluster of the M first sub-clusters, wherein M is a step that is an integer greater than or equal to 1;
obtaining a second image data set and merging the second image data set and the first cluster using the first cluster center. wherein said first cluster comprises a first cluster A, a first cluster B and a first cluster C, and merging said second image data set and said first cluster using said first cluster center comprises:
if the second image data set contains a plurality of image data, clustering the plurality of image data to obtain isolated image data and a second cluster;
merging the isolated image data into the first cluster A using the first cluster center and merging the second cluster and the first cluster B using the first cluster center;
merging said single image data into said first cluster C using said first cluster center if only a single image data exists in said second image data set. A method for incremental clustering of images as described in . The first cluster has a corresponding second cluster center, and before using the first cluster center to merge the second image data set and the first cluster,
3. The method of incremental clustering of images of claim 2, further comprising determining K first clusters from the first cluster using the second cluster centers. said second cluster having a corresponding third cluster center, and using said second cluster center to determine K first clusters from said first cluster;
obtaining a first similarity between the isolated image data and the second cluster center;
sorting the first clusters in descending order based on the first similarity to obtain a first cluster sequence, and selecting K first clusters from the top in the first cluster sequence;
obtaining a second similarity between the third cluster center and the second cluster center;
sorting the first clusters in descending order based on the second similarity to obtain a second cluster sequence, and selecting K first clusters from the top in the second cluster sequence. , or
obtaining a third similarity between the single image data and the second cluster center;
sorting the first clusters in descending order based on the third similarity to obtain a third cluster sequence, and selecting K first clusters from the top in the third cluster sequence. A method for incremental clustering of images as described in . merging the isolated image data and the first cluster A using the first cluster center,
obtaining a fourth similarity between the isolated image data and a first cluster center D, wherein the first cluster center D is the first being the first cluster center corresponding to a sub-cluster;
For each first cluster of the K first clusters, determine a first quantity of the first cluster centers D within each first cluster for which the fourth similarity is greater than a first threshold. and
determining a first cluster among the K first clusters having the largest first quantity as the first cluster A;
4. The method of incremental clustering of images of claim 3, comprising merging the isolated image data and the first cluster A. merging the second cluster and the first cluster B using the first cluster center,
dividing the second cluster into N second sub-clusters and obtaining a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters, wherein N is a step that is an integer greater than or equal to 1;
obtaining a fifth similarity between the fourth cluster center and the first cluster center E, wherein the first cluster center E is the first cluster center of each first cluster among the K first clusters; being the first cluster center corresponding to a sub-cluster;
For each first cluster of the K first clusters, determine a second quantity of the first cluster centers E within each first cluster for which the fifth similarity is greater than a second threshold. and
determining, as the first cluster B, a first cluster among the K first clusters that has the largest second quantity;
4. The method of incremental clustering of images of claim 3, comprising merging the second cluster and the first cluster B. merging the single image data and the first cluster C using the first cluster center;
obtaining a sixth similarity measure between the single image data and a first cluster center F, wherein the first cluster center F is for each first cluster of the K first clusters; being the first cluster center corresponding to a first sub-cluster;
For each first cluster of the K first clusters, determine a third quantity of the first cluster centers F within each first cluster for which the sixth similarity is greater than a third threshold. and
determining, as the first cluster C, the first cluster having the largest third quantity among the K first clusters;
4. The method of incremental clustering of images of claim 3, comprising merging the single image data and the first cluster C. wherein M is less than or equal to a fourth threshold, and after merging the second image data set and the first cluster using the first cluster center;
dividing the merged first cluster into R third sub-clusters and obtaining a fifth cluster center of each third sub-cluster of said R third sub-clusters, wherein said R is 1 a step that is an integer greater than or equal to
if the R is less than or equal to the fourth threshold, retain the R third sub-clusters and update the first cluster center using the fifth cluster center corresponding to the R third sub-clusters; a step;
obtaining a fourth quantity of image data in each third sub-cluster of the R third sub-clusters, if the R is greater than the fourth threshold;
sorting the R third sub-clusters in ascending order according to the fourth quantity to obtain a fourth cluster sequence, selecting P third sub-clusters from the top in the fourth cluster sequence; 2. updating the first cluster center with the fifth cluster center corresponding to P third sub-clusters, wherein the P is less than or equal to the fourth threshold. 8. A method for incremental clustering of images according to any one of 1 to 7. The first cluster is obtained by clustering the image data in the first image data set, and the step of dividing the first cluster into M first sub-clusters comprises:
obtaining a seventh similarity between image data in the first cluster to obtain a similarity matrix;
dividing said first cluster into said M first sub-clusters based on said similarity matrix. dividing the first cluster into the M first sub-clusters based on the similarity matrix;
obtaining a connected graph configured with the image data in the first cluster as vertices;
searching from the similarity matrix to obtain the seventh similarity between vertices of the connectivity graph;
dividing a plurality of vertices with said seventh similarity greater than a fifth threshold as one first sub-cluster to obtain said M first sub-clusters. clustering method. a first acquisition module configured to acquire a first cluster of the first image data set;
a first division configured to divide the first cluster into M first sub-clusters and obtain a first cluster center corresponding to each first sub-cluster of the M first sub-clusters; a first divided module, wherein M is an integer of 1 or more;
a merge module configured to obtain a second image data set and merge the second image data set and the first cluster using the first cluster center. The first cluster includes a first cluster A, a first cluster B and a first cluster C, and the merge module comprises:
a clustering sub-module configured to cluster the plurality of image data to obtain isolated image data and a second cluster, if the second image data set includes a plurality of image data;
a first merging sub-module configured to merge the isolated image data into the first cluster A using the first cluster center;
a second merging sub-module configured to merge the second cluster and the first cluster B using the first cluster center;
a third merge configured to merge the single image data into the first cluster C using the first cluster center if only a single image data exists in the second image data set; 12. The apparatus for incremental clustering of images according to claim 11, comprising: a sub-module. The first cluster has a corresponding second cluster center, the merge module comprising:
13. The apparatus for incremental clustering of images according to claim 12, further comprising a first determination sub-module configured to determine K first clusters from the first cluster using the second cluster centers. said second cluster having a corresponding third cluster center, said first determining sub-module comprising:
a first obtaining unit configured to obtain a first similarity between the isolated image data and the second cluster center;
A first cluster configured to sort the first clusters in descending order based on the first similarity to obtain a first cluster sequence, and select K first clusters from the top in the first cluster sequence. 1 sorting unit;
a second obtaining unit configured to obtain a second similarity measure between the third cluster center and the second cluster center;
A second cluster configured to sort the first clusters in descending order based on the second similarity to obtain a second cluster sequence, and select K first clusters from the top in the second cluster sequence. 2 sorting units, or a third obtaining unit configured to obtain a third similarity between the single image data and the second cluster centers;
A third cluster configured to sort the first clusters in descending order based on the third similarity to obtain a third cluster sequence, and select K first clusters from the top in the third cluster sequence. 14. The apparatus for incremental clustering of images according to claim 13, comprising: 3 sorting units. The first merging submodule includes:
a fourth obtaining unit configured to obtain a fourth similarity measure between the isolated image data and a first cluster center D, wherein the first cluster center D is one of the K first clusters; a fourth obtaining unit that is the first cluster center corresponding to each first sub-cluster of each first cluster;
For each first cluster of the K first clusters, determine a first quantity of the first cluster centers D within each first cluster for which the fourth similarity is greater than a first threshold. a first determining unit configured to
a second determining unit configured to determine the first cluster among the K first clusters with the largest first quantity as the first cluster A;
14. The apparatus for incremental clustering of images according to claim 13, comprising a first merging unit configured to merge the isolated image data and the first cluster A. The second merging submodule includes:
A first division configured to divide the second cluster into N second sub-clusters and obtain a fourth cluster center corresponding to each second sub-cluster of the N second sub-clusters. a first divisional unit, wherein N is an integer of 1 or more;
a fifth obtaining unit configured to obtain a fifth similarity measure between the fourth cluster center and a first cluster center E, wherein the first cluster center E is one of K first clusters a fifth obtaining unit that is the first cluster center corresponding to each first sub-cluster of each first cluster;
For each first cluster of the K first clusters, determine a second quantity of the first cluster centers E within each first cluster for which the fifth similarity is greater than a second threshold. a third determining unit configured to
a fourth determining unit configured to determine the first cluster among the K first clusters with the largest second quantity as the first cluster B;
14. Apparatus for incremental clustering of images according to claim 13, comprising a second merging unit configured to merge said second cluster and said first cluster B. The third merging submodule includes:
a sixth obtaining unit configured to obtain a sixth similarity measure between the single image data and a first cluster center F, wherein the first cluster center F is one of the K first clusters a sixth obtaining unit that is the first cluster center corresponding to each first sub-cluster of each first cluster thereof;
For each first cluster of the K first clusters, determine a third quantity of the first cluster centers F within each first cluster for which the sixth similarity is greater than a third threshold. a fifth determining unit configured to
a sixth determining unit configured to determine the first cluster among the K first clusters for which the third quantity is the largest as the first cluster C;
14. Apparatus for incremental clustering of images according to claim 13, comprising a third merging unit configured to merge said single image data and said first cluster C. The M is equal to or less than the fourth threshold,
A second splitting module configured to split the merged first cluster into R third sub-clusters and obtain a fifth cluster center of each third sub-cluster of said R third sub-clusters. a second segmented module, wherein R is an integer greater than or equal to 1;
if the R is less than or equal to the fourth threshold, retain the R third sub-clusters and update the first cluster center using the fifth cluster center corresponding to the R third sub-clusters; a first update module configured to:
a second acquisition module configured to acquire a fourth quantity of image data in each third sub-cluster of the R third sub-clusters, if the R is greater than the fourth threshold;
sorting the R third sub-clusters in ascending order according to the fourth quantity to obtain a fourth cluster sequence, selecting P third sub-clusters from the top in the fourth cluster sequence; a second update module configured to update the first cluster center using the fifth cluster center corresponding to P third sub-clusters, wherein the P is less than or equal to the fourth threshold; 18. Apparatus for incremental clustering of images according to any one of claims 11 to 17, further comprising: 2 update modules. The first cluster is obtained by clustering the image data in the first image data set, and the first division module includes:
an acquisition sub-module configured to acquire a seventh degree of similarity between image data in the first cluster to obtain a similarity matrix;
a splitting sub-module configured to split the first cluster into the M first sub-clusters based on the similarity matrix. Incremental clustering device. The divided sub-modules are
a seventh obtaining unit configured to obtain a connectivity graph configured with the image data in the first cluster as vertices;
a search unit configured to search from the similarity matrix to obtain the seventh similarity between vertices of the connectivity graph;
a second segmentation unit configured to segment a plurality of vertices with the seventh similarity greater than a fifth threshold as one first sub-cluster to obtain the M first sub-clusters. Item 20. Apparatus for incremental clustering of images according to Item 19. including input devices and output devices;
a processor suitable for implementing one or more instructions;
an electronic device, further comprising a computer storage medium loaded by said processor and storing one or more instructions suitable for carrying out the method of any one of claims 1 to 10. A computer storage medium storing one or more instructions suitable for being loaded by a processor to carry out the method of any one of claims 1 to 10. A computer program product containing one or more instructions suitable to be loaded by a processor to carry out the method of any one of claims 1 to 10.

Images