Multi-view clustering image segmentation method and system based on embedded approximate learning

Abstract

The invention discloses a multi-view clustering image segmentation method and system based on embedded approximate learning. The method comprises the following steps: acquiring an image multi-view data set which needs to be subjected to image segmentation; inputting the image multi-view data set into a pre-constructed clustering model, iteratively updating parameters of the clustering model according to an optimization target, and obtaining an optimized clustering model when a change value of the optimization target is smaller than a set threshold value; inputting the image multi-view data set into the optimized clustering model, and outputting a clustering result, thereby completing multi-view subspace clustering; and performing image segmentation according to the clustering result to obtain an image segmentation result. According to the method, self-characterization learning, a robust principal component analysis technology and Grassmann manifold space approximate learning are combined, so that the increase of calculation cost caused by eigenvalue decomposition of similar matrixes in the optimization process is avoided, and the clustering efficiency and quality are improved.

Description

Technical Field

[0001] The present invention relates to technical fields such as machine learning, data mining technology and graph optimization, and in particular to a multi-view clustering image segmentation method and system based on embedded approximate learning. Background Art

[0002] As a basic technology in unsupervised learning, clustering has promoted innovation and change in various disciplines. Its application range is wide, covering many fields such as computer vision, recommendation systems, and community detection. The core goal of clustering is to reveal the inherent patterns of data and divide data into different categories based on similarities. With the rapid development of Internet technology, how to effectively represent multi-source or heterogeneous information has become a research focus. For example, a piece of information can be presented as text, images, or documents written in multiple languages, and even individuals can be identified by fingerprints, personality traits, or voiceprints. Compared with single-view clustering methods, multi-view clustering makes full use of the complementarity between views and significantly improves the performance of clustering.

[0003] In recent years, with the continuous development of multi-view clustering algorithms, researchers have mainly used the complementarity and distinguishability between different views to improve clustering performance. Self-representation technology can effectively learn the specific information features in each view and fully explore the complementarity and distinguishability between different views. However, this learning method often relies on the self-expression matrix obtained from the original data, which may introduce outliers and noise, thereby affecting the robustness of the clustering model (J. Zhao and G.Lu, "Clean affinity matrix learning with rank equality constraint for multi-view subspace clustering," Pattern Recognit., vol. 134, p. 109118, 2023.). In order to alleviate the impact of outliers and noise, researchers usually use graph filters to smooth the feature information, which can not only maintain the inherent diversity of the input features, but also eliminate certain noise (Z. Lin, Z. Kang, L. Zhang and L.Tian, ​​"Multi-View Attributed Graph Clustering," in IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 2, pp. 1872-1880,2023.). However, most studies directly use graph filters to process the data set, which is only a preprocessing technique and is not effectively introduced into the model optimization process. Therefore, it is impossible to effectively filter out noise during model iteration, which affects the robustness of the model. To improve the robustness of the model, researchers use the Robust Principal Component Analysis (RPCA) algorithm to decompose the input matrix into a clean and rich similarity matrix and a noise matrix, thereby effectively capturing the noise in the input data (J.-B. Zhao and G.-F. Lu, “Clean and robust affinity matrix learning for multi-view clustering,” Appl. Intell., vol. 52,no. 14, pp. 15899–15915, 2022.). However, the RPCA method ignores the connection between views, resulting in a certain weakening of the clustering performance. In order to strengthen the connection between different views, the researchers proposed a clustering method that integrates different embeddings of multiple views on the Grassmann manifold from the perspective of spectral clustering, and reconstructed the similarity graph of each view using adaptive embedding, thereby maintaining the topological information and local structure in each view, further improving the accuracy and robustness of clustering (F. Qi, J. Guo, J. Li, Y. Liao, W. Liao, H. Cai, J. Chen, “Multi-kernel clustering with tensor fusion on grassmann manifold for high-dimensional genomic data”, Methods, vol. 231, pp. 215–225, 2024.).

[0004] There are many problems in the field of multi-view clustering. First, the construction of the similarity matrix is ​​affected by different measurement standards. For example, the composition based on Euclidean distance requires the number of neighbor nodes contained in the node to be adjusted, which affects the clustering performance of the model. Secondly, the original data set contains a lot of redundancy and noise, which also introduces a lot of noise during the composition. Finally, from the perspective of views, directly fusing different views ignores the local structure between each view. Summary of the invention

[0005] The present invention proposes a multi-view clustering image segmentation method and system based on embedded approximate learning, which aims to adaptively approximate learning of pure and rich similarity matrices in Grassmann manifold space, effectively eliminate the noise influence of the original data set, and improve the robustness and clustering accuracy of the model. Specifically, firstly, the self-expression matrix is ​​obtained from the original multi-view data set through the self-representation method; then the self-expression matrix is ​​decomposed using robust principal component analysis to obtain the pure similarity matrix of each view; finally, in order to strengthen the connection between each view, each similarity matrix is ​​adaptively approximate learned in the manifold space, so as to obtain the model-rich embedded features for model clustering, and image segmentation is performed according to the clustering results.

[0006] The purpose of the present invention is achieved by at least one of the following technical solutions.

[0007] A multi-view clustering image segmentation method based on embedding approximate learning includes the following steps: S1. Obtain a multi-view dataset of images that require image segmentation; S2, inputting the image multi-view dataset into a pre-built clustering model, iteratively updating the parameters of the clustering model according to the optimization target, and ending the iteration when the optimization target meets the requirements to obtain an optimized clustering model; S3, inputting the image multi-view data set into the optimized clustering model, outputting the clustering result, thereby completing the multi-view subspace clustering; S4. Perform image segmentation according to the clustering result to obtain an image segmentation result.

[0008] Furthermore, in the clustering model, self-representation technology is used to obtain a self-expression matrix from an input image multi-view dataset. In order to remove noise in the self-expression matrix, robust principal component analysis is used to decompose the self-expression matrix to obtain a pure similarity matrix. In order to strengthen the connection between each view, approximate learning is performed on the Grassmann manifold space to adaptively reconstruct the pure similarity matrix to obtain the optimal embedding, and clustering is performed using the optimal embedding to output the clustering result.

[0009] Furthermore, the image multi-view dataset is ,in, is the number of views in the image multi-view dataset, Represents the image multi-view dataset Views, is the number of samples used for clustering in each view of the image multi-view dataset, Indicates The dimension of the view; Based on the image multi-view dataset, the self-representation technology is used to obtain the self-expression matrix of each view , Represents the image multi-view dataset The self-expression matrix of the views, .

[0010] Furthermore, in order to remove the noise in the self-expression matrix, the self-expression matrix is ​​decomposed by robust principal component analysis to obtain a pure similarity matrix, as follows: Since the self-expression matrix of the view comes from the image multi-view dataset and contains a lot of noise, which affects the clustering results of the clustering model, the Robust Principal Component Analysis (RPCA) algorithm is used to decompose the self-expression matrix of each view into a pure similarity matrix and a noise matrix , , as follows: ; in, is the number of views in the image multi-view dataset, is the third regularization parameter, ; and They represent the first The similarity matrix and noise matrix of each view, , ; represents the nuclear norm of the matrix, yes norm; the RPCA algorithm uses the nuclear norm on the similarity matrix obtained by decomposition, thereby ensuring a low-rank characteristic of the similarity matrix to remove redundant information, and also uses a The norm operation makes the noise matrix sparse and improves the robustness of the model. This paper borrows the main idea of ​​RPCA to separate noise. In order to facilitate optimization, it uses to replace and , Represents the F norm of the matrix, as follows: ; ; in, and are the first regularization parameter and the second regularization parameter, respectively. , ; represents a column vector of all 1s, Represents the similarity matrix The transpose of ; Obviously, the above formula combines the idea of ​​self-representation learning and RPCA, firstly obtains the self-expression matrix of each view, and then uses the RPCA algorithm to obtain the similarity matrix and noise matrix of each self-expression matrix, and uses Constraining the similarity matrix and noise matrix ensures the compactness of the structure; Based on the image multi-view dataset Views With this view and the self-expression matrix of the view The sum of the F norms of the differences between the products of , as the component error based on the robust principal component analysis algorithm; by Similarity matrix of views The sum of the F norms as the regularized error of the similarity matrix; by The noise matrix of the view The sum of the F norms The regularized error as a noise matrix.

[0011] Furthermore, the approximate learning is performed on the Grassmann manifold space to adaptively reconstruct the pure similarity matrix and obtain the optimal embedding, as follows: In order to strengthen the connection between views in the image multi-view dataset while maintaining the topological information between views, the similarity matrices of different views are analyzed from the perspective of spectral clustering. The Laplace matrix of Perform eigenvalue decomposition to obtain orthogonal embeddings of different views , the specific process is as follows:

[0012] in, It is the first Similarity matrix of views The Laplace matrix of , compared with the similarity matrix, it contains more refined feature information; represents the trace operation of a matrix, represents the identity matrix, From the perspective of spectral clustering, it is the In order to obtain an informative embedding for the similarity matrix on the Grassmann manifold space, Adaptive reconstruction, using approximate learning from different orthogonal embeddings To get an optimal embedding , , , Represents the clustering types of image multi-view datasets; by Orthogonal embedding of views With optimal embedding The square of the projected distance between sum As the projection error; by obtaining different orthogonal embeddings With optimal embedding The minimum square of the projection distance between them ensures the optimal embedding from the perspective of the view. The optimality is as follows: .

[0013] Furthermore, in order to achieve optimal embedding Can continuously learn through weight parameters Embed the optimal With all similarity matrices Perform adaptive reconstruction to make the learned optimal embedding Similarity matrix with each view in multi-view dataset Mutually coupled and mutually influential, the specific process is as follows: ; in, represents the optimal embedding and The weight parameter for similarity matrix reconstruction is in the form of a vector. is the weight vector The elements, Represents the weight vector The 2-norm of ; the similarity matrix after adaptive reconstruction is ; The adaptive reconstruction error is indivual The similarity matrix after adaptive reconstruction is and similarity matrix The sum of the products of the F norms of the differences between them is as follows: .

[0014] Furthermore, the optimal embedding learned K-Means technique was applied to obtain clustering results.

[0015] Further, in step S2, using the image multi-view dataset views, and iteratively updates the parameters of the clustering model according to the optimization goal. The optimization goal is to traverse the image multi-view dataset Similarity matrix of views , noise matrix , the optimal embedding , and the optimal embedding and similarity matrix Reconstructed weight parameters Iterative optimization is performed to minimize the sum of the component error based on the robust principal component analysis algorithm, the regularization error of the similarity matrix, the regularization error of the noise matrix, the projection error, and the adaptive reconstruction error, as follows: ; ; in, It is a balance parameter used to accelerate the convergence speed of the model; this optimization goal embodies the innovation of the present invention from the perspective of self-characterization technology, RPCA algorithm and adaptive reconstruction of Grassmann manifold space.

[0016] Furthermore, when the change in the sum of the component error based on the robust principal component analysis algorithm, the regularization error of the similarity matrix, the regularization error of the noise matrix, the projection error, and the adaptive reconstruction error in the previous and subsequent iterative optimizations is less than the set threshold, it indicates that the clustering model has reached the optimal state, the iterative optimization is terminated, and the optimized clustering model is obtained.

[0017] Compared with the prior art, the advantages of the present invention are: The present invention proposes a multi-view clustering image segmentation method and system based on embedded approximate learning, which aims to perform adaptive approximate learning on pure and rich similarity matrices on Grassmann manifold space, eliminate the noise in the self-expression matrix due to the original data set, improve the robustness of the clustering model and the clustering accuracy, and perform image segmentation according to the clustering results. The present invention models the acquisition of the self-expression matrix as a self-representation learning process, eliminates its noise through robust principal component analysis to obtain a pure similarity matrix, and uses embedded approximate learning to reconstruct the similarity matrix on Grassmann manifold space, thereby obtaining information-rich embedded features. The present invention combines self-representation learning, robust principal component analysis technology and Grassmann manifold space approximate learning to avoid the increase in computational cost caused by eigenvalue decomposition of the similarity matrix during the optimization process, thereby improving the efficiency and quality of clustering. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The present invention is a flowchart of a multi-view clustering image segmentation method based on embedded approximate learning in an embodiment of the present invention.

[0019] Figure 2a , Figure 2b , Figure 2c and Figure 2d Schematic diagrams of the results of ablation experiments on accuracy, normalized mutual information, purity and Rand index of three clustering models in embodiments of the present invention.

[0020] Figure 3 Schematic diagram of scalability experiment results in an embodiment of the present invention.

[0021] Figure 4a and Figure 4b They are schematic diagrams of convergence experiment results on the ORL dataset and the MSRC dataset in embodiments of the present invention.

[0022] Figure 5a , Figure 5b , Figure 5c and Figure 5d They are schematic diagrams of segmentation results of the number “4” on the data set HW in the embodiments of the present invention. DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the specific implementation of the present invention is described in detail below with reference to the accompanying drawings and examples.

[0024] In one embodiment, a multi-view clustering image segmentation method based on embedding approximate learning includes the following steps: S1. Obtain a multi-view dataset of images that require image segmentation; S2, inputting the image multi-view dataset into a pre-built clustering model, wherein the clustering model uses a self-representation technique to obtain a self-expression matrix from the input image multi-view dataset; The image multi-view dataset is ,in, is the number of views in the image multi-view dataset, Represents the image multi-view dataset Views, is the number of samples used for clustering in each view of the image multi-view dataset, Indicates The dimension of the view; Based on the image multi-view dataset, the self-representation technology is used to obtain the self-expression matrix of each view , Represents the image multi-view dataset The self-expression matrix of the views, .

[0025] In order to remove the noise in the self-expression matrix, the robust principal component analysis is used to decompose the self-expression matrix to obtain a pure similarity matrix, as follows: Since the self-expression matrix of the view comes from the image multi-view dataset and contains a lot of noise, which affects the clustering results of the clustering model, the Robust Principal Component Analysis (RPCA) algorithm is used to decompose the self-expression matrix of each view into a pure similarity matrix and a noise matrix , , as follows: ; in, is the number of views in the image multi-view dataset, is the third regularization parameter, , in one embodiment, ; and They represent the first The similarity matrix and noise matrix of each view, , ; represents the nuclear norm of the matrix, yes norm; the RPCA algorithm uses the nuclear norm on the similarity matrix obtained by decomposition, thereby ensuring a low-rank characteristic of the similarity matrix to remove redundant information, and also uses a The norm operation makes the noise matrix sparse and improves the robustness of the model. This paper borrows the main idea of ​​RPCA to separate noise. In order to facilitate optimization, it uses to replace and , Represents the F norm of the matrix, as follows: ;

[0026] in, and are the first regularization parameter and the second regularization parameter, respectively. , , in one embodiment, , ; represents a column vector of all 1s, Represents the similarity matrix The transpose of ; Obviously, the above formula combines the idea of ​​self-representation learning and RPCA, firstly obtains the self-expression matrix of each view, and then uses the RPCA algorithm to obtain the similarity matrix and noise matrix of each self-expression matrix, and uses Constraining the similarity matrix and noise matrix ensures the compactness of the structure; Based on the image multi-view dataset Views With this view and the self-expression matrix of the view The sum of the F norms of the differences between the products of , as the component error based on the robust principal component analysis algorithm; by Similarity matrix of views The sum of the F norms as the regularized error of the similarity matrix; by The noise matrix of the view The sum of the F norms The regularized error as a noise matrix.

[0027] In order to strengthen the connection between each view, approximate learning is performed on the Grassmann manifold space to adaptively reconstruct the pure similarity matrix and obtain the optimal embedding, as follows: In order to strengthen the connection between views in the image multi-view dataset while maintaining the topological information between views, the similarity matrices of different views are analyzed from the perspective of spectral clustering. The Laplace matrix of Perform eigenvalue decomposition to obtain orthogonal embeddings of different views , the specific process is as follows: ; in, It is the first Similarity matrix of views The Laplace matrix of , compared with the similarity matrix, it contains more refined feature information; represents the trace operation of a matrix, represents the identity matrix, From the perspective of spectral clustering, it is the The present invention can obtain orthogonal embeddings of different views by performing eigenvalue decomposition on the Laplacian matrices of different views. In order to obtain an embedding with rich information features on the Grassmann manifold space for the similarity matrix Adaptive reconstruction, using approximate learning from different orthogonal embeddings To get an optimal embedding , , , Represents the clustering types of image multi-view datasets; by Orthogonal embedding of views With optimal embedding The square of the projected distance between sum As the projection error; by obtaining different orthogonal embeddings With optimal embedding The minimum square of the projection distance between them ensures the optimal embedding from the perspective of the view. The optimality is as follows: .

[0028] In order to achieve optimal embedding Can continuously learn through weight parameters Embed the optimal With all similarity matrices Perform adaptive reconstruction to make the learned optimal embedding Similarity matrix with each view in multi-view dataset Mutually coupled and mutually influential, the specific process is as follows: ; in, Represents the optimal embedding and The weight parameter for similarity matrix reconstruction is in the form of a vector. is the weight vector The elements, Represents the weight vector The 2-norm of ; the similarity matrix after adaptive reconstruction is ; The adaptive reconstruction error is indivual The similarity matrix after adaptive reconstruction is and similarity matrix The sum of the products of the F norms of the differences between them is as follows: .

[0029] Use the optimal embedding to perform clustering and output the clustering results.

[0030] In one embodiment, the optimal embedding K-Means technique was applied to obtain clustering results.

[0031] Using image multi-view datasets views, and iteratively updates the parameters of the clustering model according to the optimization goal. The optimization goal is to traverse the image multi-view dataset Similarity matrix of views , noise matrix , the optimal embedding , and the optimal embedding and similarity matrix Reconstructed weight parameters Iterative optimization is performed to minimize the sum of the component error based on the robust principal component analysis algorithm, the regularization error of the similarity matrix, the regularization error of the noise matrix, the projection error, and the adaptive reconstruction error, as follows: ; ； in, is a balance parameter used to speed up the convergence of the model. In one embodiment, ; This optimization goal embodies the innovation of the present invention from the perspectives of self-characterization technology, RPCA algorithm and adaptive reconstruction of Grassmann manifold space.

[0032] When the change value of the sum of the component error based on the robust principal component analysis algorithm, the regularization error of the similarity matrix, the regularization error of the noise matrix, the projection error, and the adaptive reconstruction error in the previous and subsequent iterative optimizations is less than the set threshold, it indicates that the clustering model has reached the optimal state, and the iterative optimization is terminated to obtain the optimized clustering model. In one embodiment, the set threshold is 0.0001.

[0033] S3, inputting the image multi-view data set into the optimized clustering model, outputting the clustering result, thereby completing the multi-view subspace clustering; S4. Perform image segmentation based on the clustering results (see H. Zhang, X. Lu, P. Ma, J. Liu, J.Lian and Y. Ma, " Cluster fusion based cross teaching for semi-supervisedmedical image segmentation", Neurocomputing, vol. 618, pp. 129147, 2025.) to obtain the image segmentation result.

[0034] In one embodiment, the advantages of the clustering model are reflected by the performance of the accuracy (ACC), normalized mutual information (NMI), purity (P) and adjusted Rand Index (ARI) on 8 commonly used image multi-view datasets (HW, ORL, COIL20, 100Leaves, YaleB, MSRC, Scene and ALOI-100). Specifically, the present invention first illustrates the advantages of the robust principal component analysis technology based on self-representation technology and Grassmann manifold space approximate learning in the clustering model through ablation experiments, and then illustrates the high efficiency of the clustering of the present invention through scalability experiments, thereby reflecting its application value, and finally illustrates its theoretical value through the convergence experiment of the clustering model.

[0035] In the ablation experiment, in order to reflect the role of robust principal component analysis technology in removing noise, the first clustering model J1 is constructed as follows:

[0036] ; In the first clustering model J1, and are the first regularization parameter and the second regularization parameter, respectively. , In one embodiment, take , ; represents a column vector of all 1s, Represents the self-expression matrix The transpose of Represented by different orthogonal embeddings An optimal embedding is obtained from is the weight vector The elements, Represents the optimal embedding and Self-expression matrix The reconstructed weight parameter is in the form of a vector. Represents the weight vector 2 norm of; through the weight parameter To adaptively reconstruct all self-expression matrices so that the learned optimal embedding Compared with the individual views in the multi-view dataset Mutually coupled and mutually influential. The robust principal component analysis module is removed here and the self-expression matrix is ​​used directly and the optimal embedding produced on the Grassmann manifold space Perform adaptive reconstruction learning.

[0037] In order to reflect the close connection between different views in Grassmann manifold space, the second clustering model J2 is constructed as follows:

[0038] ; ; In the second clustering model J2, and are the first regularization parameter and the second regularization parameter, respectively. , ; is a balance parameter, , is the fourth regularization parameter, In one embodiment, take , , and ; represents a column vector of all 1s, Represents the self-expression matrix The transpose of Represented by all similar matrices A fused similarity matrix for adaptive fusion, is the weight vector The elements, express and Similarity Matrix The reconstructed weight parameter is in the form of a vector. Represents the weight vector 2 norm of; through the weight parameter To adaptively reconstruct all similar matrices, so that the fusion similarity matrix Similarity matrix with each view in multi-view dataset Mutually coupled and mutually influential. Here, the module of approximate learning in Grassmann manifold space is removed, and the similarity matrix obtained by robust principal component analysis technology is directly Fusion is performed, so the fused similarity matrix is ​​obtained , applying a K-Means to the fused similarity matrix can get the clustering result.

[0039] The results of the ablation experiment are as follows Figure 2a , Figure 2b, Figure 2c and Figure 2d shown.

[0040] From the results of the ablation experiment, it can be seen that the clustering model of the present invention is based on the same evaluation index, and only has a slightly lower accuracy than the first clustering model J1 on the data set ALOI-100, and is better than the first clustering model J1 and the second clustering model J2 in the rest of the cases. Overall, the clustering model of the present invention introduces robust principal component analysis technology and Grassmann manifold space approximate learning, which has obvious advantages in clustering.

[0041] In one embodiment, a scalability experiment is performed. In order to reflect the high efficiency of the present invention in clustering, the ALOI-100 dataset is selected to calculate its running time under different input ratios: the ALOI-100 dataset is divided into 10 parts on average, and different numbers of parts are randomly input each time. Then, the time of different input ratios is fitted to verify its high efficiency.

[0042] The scalability experiment results are as follows Figure 3 As shown, from Figure 3 It can be seen from the fitting curve in that: for the running time of different input ratios of the ALOI-100 data set, the fitting curve of the clustering model of the present invention can well satisfy the distribution of the quadratic function, which shows that the clustering model of the present invention has extremely high application value.

[0043] In one embodiment, the convergence of the clustering model can ensure that the model optimization can obtain an optimal result, thereby proving the theoretical significance of the model. The present invention selects the data set ORL and the data set MSRC to perform convergence experiments to prove the convergence. The convergence curves of the two data sets are shown in Figure 2. Figure 4a and Figure 4b shown.

[0044] From the convergence of the clustering model of the present invention on the data set ORL and the data set MSRC: as the number of iterations increases, the target value decreases rapidly and monotonically until it finally approaches a fixed value. This shows that the clustering model of the present invention has good convergence and has scientific theoretical significance.

[0045] In one embodiment, the HW data set is analyzed based on the clustering model result. Figure 5a The number "4" shown in the figure is segmented, and the present invention has great prospects in image processing by adjusting different cluster centers. The segmentation result of the number "4" is as follows Figure 5b and Figure 5c In order to demonstrate the segmentation effect of the present invention on image data, the segmentation results generated by the region growing algorithm are compared, as shown in FIG. Figure 5dAs shown in the figure, it can be seen from the comparison that the segmented image generated by the present invention can segment the slight changes in the same area, and the effect is better.

[0046] The preferred embodiments of the present application disclosed above are only used to help understand the present invention and its core ideas. For those skilled in the art, according to the ideas of the present invention, there will be changes in specific application scenarios and implementation operations, and this specification should not be interpreted as limiting the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A multi-view clustering image segmentation method based on embedding approximate learning, characterized in that: The steps include: S1. Obtain a multi-view dataset of images that require image segmentation; S2, inputting the image multi-view dataset into a pre-built clustering model, iteratively updating the parameters of the clustering model according to the optimization target, and ending the iteration when the optimization target meets the requirements to obtain an optimized clustering model; S3, inputting the image multi-view data set into the optimized clustering model, outputting the clustering result, thereby completing the multi-view subspace clustering; S4. Perform image segmentation according to the clustering result to obtain an image segmentation result.

2. The multi-view clustering image segmentation method based on embedded approximate learning according to claim 1, characterized in that: In the clustering model, a self-expression matrix is ​​obtained from an input image multi-view dataset using a self-representation technique; a similarity matrix is ​​obtained by decomposing the self-expression matrix using a robust principal component analysis; Approximate learning is performed on the Grassmann manifold space to adaptively reconstruct the similarity matrix and obtain the optimal embedding. The optimal embedding is used for clustering and the clustering result is output.

3. The multi-view clustering image segmentation method based on embedded approximate learning according to claim 2, characterized in that: The image multi-view dataset is ,in, is the number of views in the image multi-view dataset, Represents the image multi-view dataset Views, , is the number of samples used for clustering in each view of the image multi-view dataset, Indicates The dimension of the view; Based on the image multi-view dataset, the self-representation technology is used to obtain the self-expression matrix of each view , Represents the image multi-view dataset The self-expression matrix of the views, .

4. The multi-view clustering image segmentation method based on embedded approximate learning according to claim 2, characterized in that: The robust principal component analysis algorithm is used to decompose the self-expression matrix of each view into a similarity matrix and a noise matrix; Based on the image multi-view dataset The sum of the F-norms of the differences between a view and the product of the view and the self-expression matrix of the view is used as the component error based on the robust principal component analysis algorithm; by The sum of the F norms of the similarity matrices of the views is taken as the regularization error of the similarity matrix; by The sum of the F-norms of the noise matrices of the views is taken as the regularized error of the noise matrix.

5. The multi-view clustering image segmentation method based on embedded approximate learning according to claim 2, characterized in that: The approximate learning is performed on the Grassmann manifold space to adaptively reconstruct the pure similarity matrix and obtain the optimal embedding, as follows: From the perspective of spectral clustering, the Laplacian matrix of the similarity matrix of different views is subjected to eigenvalue decomposition to obtain the orthogonal embedding of different views; Use approximate learning to obtain an optimal embedding from different orthogonal embeddings; The projection error is taken as the sum of the squares of the projected distances between the orthogonal embedding of the view and the optimal embedding.

6. The multi-view clustering image segmentation method based on embedding approximate learning according to claim 5, characterized in that: The optimal embedding is obtained by weighting the Adaptive reconstruction with all similarity matrices; by indivual The similarity matrix after adaptive reconstruction is and similarity matrix The sum of the products of the F norms of the differences between them is taken as the adaptive reconstruction error; Represents the optimal embedding and Similarity Matrix The reconstructed weight parameter is in the form of a vector. is the weight vector The elements; the similarity matrix after adaptive reconstruction is .

7. The multi-view clustering image segmentation method based on embedded approximate learning according to claim 1, characterized in that: K-Means technique is applied to the optimal embedding to obtain clustering results.

8. The multi-view clustering image segmentation method based on embedded approximate learning according to any one of claims 2 to 6, characterized in that: In step S2, the optimization goal is to traverse the image multi-view dataset Similarity matrix, noise matrix, optimal embedding of views , and the optimal embedding The weight parameters of the similarity matrix reconstruction are iteratively optimized to minimize the sum of the component error based on the robust principal component analysis algorithm, the regularization error of the similarity matrix, the regularization error of the noise matrix, the projection error and the adaptive reconstruction error.

9. The multi-view clustering image segmentation method based on embedded approximate learning according to claim 8, characterized in that: When the change in the sum of the component error based on the robust principal component analysis algorithm, the regularization error of the similarity matrix, the regularization error of the noise matrix, the projection error, and the adaptive reconstruction error in the previous and subsequent iterative optimizations is less than the set threshold, it indicates that the clustering model has reached the optimal state, the iterative optimization is terminated, and the optimized clustering model is obtained.

10. A multi-view clustering image segmentation system based on embedded approximate learning that implements the method of claim 1, characterized in that: Includes the following modules: Data acquisition module: obtains the multi-view image dataset required for image segmentation; Clustering model optimization module: input the image multi-view dataset into the pre-built clustering model, iteratively update the parameters of the clustering model according to the optimization target, and obtain the optimized clustering model when the change value of the optimization target is less than the set threshold; Clustering module: inputting the image multi-view data set into the optimized clustering model and outputting the clustering result, thereby completing the multi-view subspace clustering; Image segmentation module: Perform image segmentation based on the clustering results to obtain image segmentation results.

Images