The Application Of Sparse Subspace Clustering In Image Processing

With the rapid development of science and technology,we are already in an era of big data explosion.Many problems in the real world involve collections of high-dimensional data such as images,videos,web documents,DNA microarray data and more.These high-dimensional data not only increase the running time and memory requirements of the algorithm but also cause some difficulties in handling the problems due to the noise pollution and the shortage of samples.Fortunately,the structure of high-dimensional data is not without any rules at all,they are usually among distributed in a number of low dimensional subspace and.Using subspace clustering,we can find out the spatial pattern of spatial entities,reveal the distribution rules of spatial entities,extract the spatial structure features of spatial entities,and predict the trends of spatial entities.In recent years,Low-Rank Representation(LRR)has been widely used in face recognition,image clustering,motion segmentation and feature extraction,and has made remarkable achievements in exploring low-dimensional subspace structures.For a given data set containing sparse errors,the purpose of the LRR is to find the lowest common expression of all data.In the real world,data is often stored on low-dimensional submanifolds in high-dimensional environmental space.However,the low rank of LRR indicates that the model deviates from the rank function and does not consider the inherent geometry of the data.High-dimensional data clustering has low accuracy.In view of the above problems,this paper presents a kind of Laplacian Regularized Hyperbolic Tangent Function Low-Rank Representation model(LRHT-LRSC)based on a comprehensive analysis of the low-rank representation model and the internal geometry of the data.The main work in this paper includes the following aspects:A kind of Laplacian Regularized Hyperbolic Tangent Function Low-Rank Representation model(LRHT-LRSC)is proposed.The LRHT-LRSC could compact approximation to the rank by using a hyperbolic tangent function instead of the nuclear norm and the accuracy of data clustering be improved by utilized Laplacian Regularizer taking into account the intrinsic geometrical structures within the data.Then,the coefficient matrix and the similarity matrix of the data samples are constructed.Finally,the final clustering results are obtained by using the spectral clustering method.Experimental results on synthetic datasets,real data Extended Yale B and Hopkins 155 show that the proposed LRHT-LRR improves the accuracy and robustness of clustering.Although the LRHT-LRSC algorithm reduces the error rate of high-dimensional data clustering,it improves the robustness of clustering algorithm but improves the run-time time of the algorithm.To solve the problem and based on the LRHT-LRSC algorithm,the distributed and multi-core parallel computation of Matlab is fully utilized deal with the slow and time-consuming problem of serial computation,and then realize multi-core parallelization of LRHT-LRSC.The experimental results show that compared with the traditional method,the multi-core parallel method can effectively reduce the clustering time under the premise of ensuring the accuracy and stability.