Multi-view clustering method, terminal device, and computer-readable storage medium

Abstract

The application provides a multi-view clustering method, a terminal device and a computer readable storage medium. The method comprises the following steps: obtaining an anchor point graph matrix of each view data based on a multi-view data matrix; splicing the anchor point graph matrix of each view data to obtain a fusion specific anchor point graph matrix; obtaining a shared anchor point graph matrix of the multi-view data based on the multi-view data matrix; splicing the fusion specific anchor point graph matrix and the shared anchor point graph matrix to obtain a fusion anchor point graph matrix; obtaining a fusion feature matrix of the multi-view data based on the fusion anchor point graph matrix; and clustering the fusion feature matrix by using a preset clustering algorithm to obtain a clustering result of the multi-view data. Compared with a multi-view clustering method based on an anchor point graph and a representative incomplete multi-view clustering method, the method provided by the application has a faster calculation speed; in addition, the exploration of specific information and shared information of the multi-view data and the fusion of the specific information and the shared information make the clustering precision of the method higher.

Description

Technical Field

[0001] This application relates primarily to the field of multi-view image technology, and in particular to a multi-view clustering method, a terminal device, and a computer-readable storage medium. Background Technology

[0002] Anchor graphs are an effective method for constructing large-scale graph matrices and have demonstrated excellent performance in large-scale multi-view data spectral clustering tasks. However, no work has yet investigated the application of anchor graphs in incomplete multi-view data clustering problems. Current incomplete multi-view data clustering algorithms generally have high computational complexity on large-scale data and have not considered the problem of discovering and integrating private information of each view and shared information between views. Summary of the Invention

[0003] This application provides a multi-view clustering method, a terminal device, and a computer-readable storage medium.

[0004] To address the aforementioned technical problems, this application provides a multi-view clustering method, which includes:

[0005] Retrieves a multi-view data matrix containing multi-view data;

[0006] Based on the multi-view data matrix, obtain the anchor point map matrix for each view data;

[0007] The anchor point map matrix of each view data is concatenated to obtain a fused specific anchor point map matrix;

[0008] Based on the multi-view data matrix, obtain the shared anchor point map matrix of the multi-view data;

[0009] By splicing the fused specific anchor point map matrix and the shared anchor point map matrix, a fused anchor point map matrix is ​​obtained;

[0010] Based on the fusion anchor point map matrix, obtain the fusion feature matrix of the multi-view data;

[0011] The fused feature matrix is ​​clustered using a preset clustering algorithm to obtain the clustering results of the multi-view data.

[0012] The step of obtaining the anchor point map matrix for each view data based on the multi-view data matrix includes:

[0013] Based on the multi-view data matrix, obtain the data for each view;

[0014] A number of samples are sampled from each view data to form the anchor point matrix of the corresponding view data;

[0015] Using the anchor point matrix, obtain the anchor point graph matrix for each view data.

[0016] The step of obtaining the anchor point map matrix for each view data using the anchor point matrix includes:

[0017] The first objective function between view data and view anchors is constructed using the anchor matrix.

[0018] The anchor point graph matrix is ​​obtained by minimizing the first objective function.

[0019] After obtaining the anchor point matrix for each view data using the anchor point matrix, the multi-view clustering method further includes:

[0020] The anchor point matrix of each view data is column normalized as follows:

[0021]

[0022] Among them, Z v The anchor point graph matrix before column normalization, This is the normalized anchor point graph matrix.

[0023] The step of obtaining the shared anchor point map matrix of the multi-view data based on the multi-view data matrix includes:

[0024] Initialize the anchor point matrix for each view data, and initialize the shared anchor point graph matrix;

[0025] Construct a second objective function relating the anchor point matrix and the shared anchor point graph matrix;

[0026] The multi-view data matrix is ​​input into the second objective function, and the anchor point matrix of each view data and the shared anchor point graph matrix of the multi-view data are obtained through iterative solution.

[0027] The step of splicing the fused specific anchor point map matrix and the shared anchor point map matrix to obtain the fused anchor point map matrix includes:

[0028] Obtain the balance parameter between the fused specific anchor point map matrix and the shared anchor point map matrix;

[0029] The fused anchor map matrix is ​​obtained by splicing the fused specific anchor map matrix and the shared anchor map matrix using the balancing parameters.

[0030] The step of obtaining the fusion feature matrix of the multi-view data based on the fusion anchor point map matrix includes:

[0031] Singular value decomposition is performed on the fused anchor point map matrix to obtain the fused feature matrix of the multi-view data composed of left singular vectors.

[0032] The multi-view clustering method further includes, after clustering the fused feature matrix using a preset clustering algorithm to obtain the clustering results of the multi-view data:

[0033] Based on the clustering results of the multi-view data, a prediction sample set is obtained, wherein the prediction sample set includes: the number of positive samples correctly predicted as positive samples, the number of negative samples correctly predicted as negative samples, the number of negative samples incorrectly predicted as positive samples, and the number of positive samples incorrectly predicted as negative samples.

[0034] Based on the predicted sample set, a clustering result evaluation index is calculated to evaluate the clustering effect of the multi-view data.

[0035] The evaluation metrics for the clustering results are accuracy, precision, recall, and / or F-score.

[0036] To solve the above-mentioned technical problems, this application provides a terminal device, wherein the terminal device includes a processor and a memory connected to the processor, wherein the memory stores program instructions;

[0037] The processor is used to execute program instructions stored in the memory to implement the multi-view clustering method described above.

[0038] To address the aforementioned technical problems, this application provides a computer-readable storage medium storing program instructions that, when executed, implement the aforementioned multi-view clustering method.

[0039] Compared with existing technologies, the beneficial effects of this application are as follows: the terminal device acquires a multi-view data matrix of multi-view data; based on the multi-view data matrix, it acquires an anchor point map matrix for each view data; it concatenates the anchor point map matrices of each view data to obtain a fused specific anchor point map matrix; based on the multi-view data matrix, it acquires a shared anchor point map matrix of the multi-view data; it concatenates the fused specific anchor point map matrix and the shared anchor point map matrix to obtain a fused anchor point map matrix; based on the fused anchor point map matrix, it acquires a fused feature matrix of the multi-view data; and it uses a preset clustering algorithm to cluster the fused feature matrix to obtain the clustering result of the multi-view data. Compared with similar anchor point map-based multi-view clustering methods and representative incomplete multi-view clustering methods, the method proposed in this application has a faster computation speed; furthermore, the exploration in the discovery and fusion of specific and shared information in multi-view data makes the clustering accuracy of the proposed method higher. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0041] Figure 1 This is a flowchart illustrating an embodiment of the multi-view clustering method provided in this application;

[0042] Figure 2 This is the flowchart of the incomplete multi-view clustering implementation scheme SSAGLIMC based on the complementary information anchor graph in this application;

[0043] Figure 3 This is a schematic diagram of the framework of an embodiment of the terminal device provided in this application;

[0044] Figure 4 This is a schematic diagram of the structure of an embodiment of the computer storage medium provided in this application. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0046] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0047] To address the aforementioned issues, this application proposes a fast incomplete multi-view clustering algorithm based on complementary anchor graphs. This method first learns the private information of each view by constructing specific anchor graphs for each incomplete view. Secondly, it constructs multi-view clustering... Figure 1By using anchor point graphs to learn shared information between views and integrating private and shared information from multi-view data, the model can capture the implicit representational information within multi-view data more quickly and comprehensively.

[0048] Please refer to the details. Figure 1 and Figure 2 , Figure 1 This is a flowchart illustrating an embodiment of the multi-view clustering method provided in this application. Figure 2 This is the flowchart of the incomplete multi-view clustering implementation scheme SSAGLIMC based on the complementary information anchor graph in this application.

[0049] like Figure 1 As shown, the multi-view clustering method in this embodiment specifically includes the following steps:

[0050] Step S11: Obtain the multi-view data matrix of the multi-view data.

[0051] In this embodiment, the multi-view data may include multiple views captured from different perspectives of the same area, and may also include anchor point maps generated based on each view. The multi-view data matrix may include the original images of the multiple views, or may include anchor point maps of the multiple views, or may include both original images and anchor point maps.

[0052] Specifically, the data matrix of the v-th view in the original multi-view data is represented as follows: v = 1, 2, ..., V, where V is the number of views, m v Let n be the dimension of the v-th view sample. v Let v be the number of samples in the v-th view.

[0053] Step S12: Based on the multi-view data matrix, obtain the anchor point map matrix for each view data.

[0054] In this embodiment, the terminal device extracts each view data from the multi-view data matrix, and then samples L samples from each view data to form the anchor point matrix of the corresponding view data. Then, the terminal device uses the anchor point matrix to obtain the anchor point map matrix for each view data.

[0055] Specifically, the terminal device obtains the anchor point graph matrix Z between the view data and the view anchor points by minimizing the following objective function. v The matrix has a size of n. v ×L:

[0056]

[0057] in, For view data, This is the view anchor point.

[0058] Furthermore, the terminal device further constrains each row of the anchor point graph matrix to have only k non-zero elements, as shown in the following equation, solved row by row.

[0059]

[0060] in,

[0061] The terminal device defines a corresponding missing indicator matrix G based on the location of the missing sample in each view. v If the i-th sample in the complete data is the j-th sample in the missing data, then otherwise

[0062] Terminal devices can also normalize column normalization by standardizing the anchor point matrix of each view data in the following ways:

[0063]

[0064] Among them, Z v The anchor point graph matrix before column normalization, This is the normalized anchor point graph matrix.

[0065] Step S13: Concatenate the anchor point map matrices of each view data to obtain a fused specific anchor point map matrix.

[0066] In this embodiment of the application, the terminal device splices the anchor point map matrices of each view to obtain a fused specific anchor point map matrix in the following manner:

[0067]

[0068] in, Indicates the matrix The number of rows is determined by n v Expanding to n, for the v-th view, if the j-th sample in the missing data is the i-th sample in the complete data, i.e. So The j-th row is filled into the i-th row; otherwise, the row is filled with 0.

[0069] Step S14: Based on the multi-view data matrix, obtain the shared anchor point map matrix of the multi-view data.

[0070] In this embodiment of the application, in order to learn the shared anchor point graph matrix among multiple incomplete multi-view data, such as Figure 2 As shown, the terminal device can initialize the view anchor point matrix B. v v = 1, 2, ... V, and initialize the shared anchor point graph matrix z as a matrix with 0 elements.

[0071] Then, the terminal device minimizes the following objective function:

[0072]

[0073] Among them, L s To determine the number of anchor points corresponding to the shared anchor point map, Z = [z ij Constrain the columns of the anchor point matrix of each view to be orthogonal, i.e. (B v The matrix size is m v ×L s This is to remove information redundancy between anchor points.

[0074] The terminal device iteratively solves the anchor point matrix B of each view. v And the shared anchor point graph matrix Z (matrix size n×L) s The specific steps are as follows:

[0075] Fixed Z (t) Find B v(t+1) (v = 1, 2, ..., V), for X vT G vT Z (t) Singular value decomposition yields the left and right singular vector matrices U and V, then the anchor point matrix for each view is B. v(t+1) =UV;

[0076] Fixed B v(t+1) (v = 1, 2, ... V), calculate row by row.

[0077]

[0078] in, Among them, G v D v(t+1) This indicates that matrix D v(t+1) The number of rows is determined by n v Expanding to n, for the v-th view, if the j-th sample in the missing data is the i-th sample in the complete data, i.e. Then D v(t+1) The j-th row is filled into the i-th row; otherwise, the row is filled with 0.

[0079] Terminal devices can also standardize the shared anchor point graph matrix in the following ways:

[0080]

[0081] Step S15: Combine and merge the specific anchor point map matrix and the shared anchor point map matrix to obtain the merged anchor point map matrix.

[0082] In this embodiment, the terminal device weightedly stitches and fuses the specific anchor point map matrix and the shared anchor point map matrix to obtain the fused anchor point map matrix:

[0083]

[0084] Where τ is the balance parameter between the specific anchor graph and the shared anchor graph.

[0085] Step S16: Based on the fusion anchor point map matrix, obtain the fusion feature matrix of the multi-view data.

[0086] In this embodiment of the application, the terminal device fuses the anchor point map matrix Z. fuse Perform singular value decomposition; the matrix formed by the left singular vectors is the fusion feature matrix U. fuse .

[0087] Step S17: Use a preset clustering algorithm to cluster the fused feature matrix to obtain the clustering results of the multi-view data.

[0088] In this embodiment of the application, the terminal device uses the k-means algorithm to fuse the feature matrix U fuse Clustering. In other embodiments, other clustering algorithms may also be used to cluster the fused feature matrix U. fuse Clustering will not be discussed further here.

[0089] Furthermore, the terminal device can also obtain a prediction sample set based on the clustering results of the multi-view data, wherein the prediction sample set includes: the number of positive samples correctly predicted as positive samples, the number of negative samples correctly predicted as negative samples, the number of negative samples incorrectly predicted as positive samples, and the number of positive samples incorrectly predicted as negative samples.

[0090] The terminal device calculates clustering result evaluation metrics based on the predicted sample set to evaluate the clustering effect of the multi-view data. These evaluation metrics include accuracy, precision, recall, and / or the F-score.

[0091] For example, terminal devices can calculate clustering result evaluation metrics: clustering accuracy (AC) and F-score.

[0092]

[0093]

[0094] Where TP represents the number of positive samples correctly predicted as positive samples, TN represents the number of negative samples correctly predicted as negative samples, FP represents the number of negative samples incorrectly predicted as positive samples, and FN represents the number of positive samples incorrectly predicted as negative samples. Ten clustering operations are performed, and the mean of the ten clustering results is recorded.

[0095] This application includes experimental verification: the proposed method is compared with several popular algorithms in terms of clustering performance (clustering accuracy AC and F-score) and computational efficiency on incomplete multi-view data. The results are shown in Table 1. The multi-view data is a dataset of multi-source dermoscopic images of common pigmented skin lesions, containing 10015 samples. Pixel features, Gabor wavelet features, LBP features, and grayscale histogram features are extracted as four views of the data, with corresponding feature dimensions of 784, 392, 256, and 11, respectively. Each view has random missing samples, with a missing sample ratio of 30%. MVSC, LMVSC, BMVC, and SMVSC are complete multi-view clustering algorithms. For these four algorithms, the missing samples are first filled with the average value of each view, and then the data is used as the complete multi-view data input to the algorithm. For all algorithms, the runtime test does not include the time consumed by the k-means clustering process in the last step; only the time consumed by each algorithm to generate the final fused feature matrix is ​​considered. All methods were tested for computational complexity on the same computer configuration: Windows 10 system, MATLAB 2017a, Intel Core 9700 CPU, and 16GB RAM. It can be seen that the algorithm in this application significantly outperforms the comparison algorithms in clustering performance and also has the highest computational efficiency.

[0096] Table 1 compares the clustering performance and computation time of several current anchor point-based multi-view clustering algorithms and representative incomplete multi-view clustering algorithms on multi-view data with a 30% missing data ratio.

[0097]

[0098] In this embodiment, the terminal device acquires a multi-view data matrix of multi-view data; based on the multi-view data matrix, it acquires an anchor point map matrix for each view data; it concatenates the anchor point map matrices of each view data to obtain a fused specific anchor point map matrix; based on the multi-view data matrix, it acquires a shared anchor point map matrix of the multi-view data; it concatenates the fused specific anchor point map matrix and the shared anchor point map matrix to obtain a fused anchor point map matrix; based on the fused anchor point map matrix, it acquires a fused feature matrix of the multi-view data; and it clusters the fused feature matrix using a preset clustering algorithm to obtain the clustering result of the multi-view data. Compared with similar anchor point map-based multi-view clustering methods and representative incomplete multi-view clustering methods, the method proposed in this application has a faster computation speed; furthermore, the exploration in the discovery and fusion of specific and shared information in multi-view data makes the clustering accuracy of the proposed method higher.

[0099] This application proposes an incomplete multi-view clustering algorithm SSAGLIMC based on complementary anchor graphs: (1) Anchor points are obtained by randomly sampling the data of each view, and a special anchor graph of each view data and anchor points is constructed to learn the private information of each view; (2) A shared anchor graph learning model is designed to learn the shared information between views. During the learning process of the shared anchor graph, the model adaptively learns the anchor points of each view; (3) The special anchor graphs of each view and the shared anchor graph are spliced ​​together to obtain a fused anchor graph; (4) Singular value decomposition is performed on the fused anchor graph to obtain the fused feature representation.

[0100] To implement the multi-view clustering method in the above embodiments, this application also provides a terminal device 300, for details please refer to [link to details]. Figure 3 The terminal device 300 in this application embodiment includes a processor 31, a memory 32, an input / output device 33, and a bus 34.

[0101] The processor 31, memory 32, and input / output device 33 are respectively connected to the bus 34. The memory 32 stores program data, and the processor 31 is used to execute the program data to implement the multi-view clustering method described in the above embodiments.

[0102] In this embodiment, processor 31 can also be referred to as a CPU (Central Processing Unit). Processor 31 may be an integrated voltage control system chip with signal processing capabilities. Processor 31 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor, or processor 31 can be any conventional processor.

[0103] This application also provides a computer storage medium; please refer to the following: Figure 4 , Figure 4 This is a schematic diagram of a computer storage medium according to an embodiment of the present application. The computer storage medium 400 stores program data 41, which is used to implement the multi-view clustering method of the above embodiment when executed by the processor.

[0104] When the embodiments of this application are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A multi-view clustering method, characterized by, The multi-view clustering method includes: A multi-view data matrix is ​​obtained from multi-view data; the multi-view data is image data. Based on the multi-view data matrix, obtain the anchor point map matrix for each view data; The missing indicator matrix for each view is determined based on the location of the missing samples in each view. For each view, the number of rows in the anchor map matrix of that view is expanded to the number of samples in the complete data using the missing indicator matrix of that view, and the rows corresponding to the missing samples are filled with 0. The anchor map matrices of each view data after the row expansion are concatenated to obtain the fused specific anchor map matrix. Based on the multi-view data matrix, obtaining the shared anchor point graph matrix of the multi-view data includes: initializing the anchor point matrix of each view data and the shared anchor point graph matrix; constructing a second objective function regarding the anchor point matrix and the shared anchor point graph matrix, wherein the second objective function constrains the columns of the anchor point matrix of each view data to be orthogonal; obtaining the anchor point matrix of each view data and the shared anchor point graph matrix by iteratively solving the second objective function, wherein each iteration includes: fixing the shared anchor point graph matrix and updating the anchor point matrix of each view data; fixing the updated anchor point matrix of each view data and updating the shared anchor point graph matrix row by row; The process of concatenating the fused specific anchor graph matrix and the shared anchor graph matrix to obtain a fused anchor graph matrix includes: obtaining a balance parameter between the fused specific anchor graph matrix and the shared anchor graph matrix; weighting the fused specific anchor graph matrix using the square root of the balance parameter, and weighting the shared anchor graph matrix using the square root of the difference between 1 and the balance parameter; and concatenating the weighted fused specific anchor graph matrix and the weighted shared anchor graph matrix to obtain the fused anchor graph matrix. Based on the fusion anchor point map matrix, obtain the fusion feature matrix of the multi-view data; The fused feature matrix is ​​clustered using a preset clustering algorithm to obtain the clustering results of the multi-view data.

2. The multi-view clustering method according to claim 1, characterized in that, The step of obtaining the anchor point map matrix for each view data based on the multi-view data matrix includes: Based on the multi-view data matrix, obtain the data for each view; A number of samples are sampled from each view data to form the anchor point matrix of the corresponding view data; Using the anchor point matrix, obtain the anchor point graph matrix for each view data.

3. The multi-view clustering method according to claim 2, characterized in that, The step of obtaining the anchor point map matrix for each view data using the anchor point matrix includes: The first objective function between view data and view anchors is constructed using the anchor matrix. The anchor point graph matrix is ​​obtained by minimizing the first objective function.

4. The multi-view clustering method according to claim 2 or 3, characterized in that, After obtaining the anchor point matrix for each view data using the anchor point matrix, the multi-view clustering method further includes: The anchor point matrix of each view data is column normalized as follows: ， wherein, is the anchor graph matrix before column normalization, is the anchor graph matrix after column normalization.

5. The multi-view clustering method according to claim 1, characterized in that, The step of obtaining the fusion feature matrix of the multi-view data based on the fusion anchor point map matrix includes: Singular value decomposition is performed on the fused anchor point map matrix to obtain the fused feature matrix of the multi-view data composed of left singular vectors.

6. The multi-view clustering method according to claim 1, characterized in that, After clustering the fused feature matrix using a preset clustering algorithm to obtain the clustering results of the multi-view data, the multi-view clustering method further includes: Based on the clustering results of the multi-view data, a prediction sample set is obtained, wherein the prediction sample set includes: the number of positive samples correctly predicted as positive samples, the number of negative samples correctly predicted as negative samples, the number of negative samples incorrectly predicted as positive samples, and the number of positive samples incorrectly predicted as negative samples. Based on the predicted sample set, a clustering result evaluation index is calculated to evaluate the clustering effect of the multi-view data. The evaluation metrics for the clustering results are accuracy, precision, recall, and / or F-score.

7. A terminal device, characterized in that, The terminal device includes a processor and a memory connected to the processor, wherein... The memory stores program instructions; The processor is used to execute program instructions stored in the memory to implement the multi-view clustering method as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The storage medium stores program instructions that, when executed, implement the multi-view clustering method as described in any one of claims 1 to 6.

Images