Research On Multi-view Canonical Correlation Subspace Clustering Algorithm

With the development of transducer skill and analyze data technique,and there are many kinds of data Collection methods or a lot of feature representation ways.Such as multi-view representation of data.The core problem is how to construct the uniformity and Complementary characteristic of multi-view data,and realize the uniformity representation and comprehensive analysis of multi-view characteristics.First of all,the definitions of Canonical Correlation Analysis,Multi-view Canonical Correlation Analysis,K-means,Multi-view K-Means,Multi-view Spectral are given.Secondly,to solve the problem that Multi-view K-means and Multi-view Spectral Clustering can not effectively mine the potential subspace structure and discriminant structure of multi-view data,and Multi-view Canonical Correlation Subspace Clustering(Mv CCSC)is proposed.Based on the idea of Canonical Correlation Analysis,and Spectral Clustering and collaborative training,and Mv CCSC models the similarity among samples on different views by using self-similarity model,and improves the performance of multi-view clustering by using collaborative training approach to share information among different similarity matrices.Finally,a Multi-view collaborative clustering approach based on Kernel Canonical Correlation Analysis Spectral Clustering(KCCAMv SC)presume is suggest and discussed to explore a consistent solution from plural representations and improve the performance of classical clustering systems by using the message in each representations.In order to verify Mv CCSC and KCCAMv SC,The results of the experiments were carried out on image and text data sets with different sizes and features.The results of the experiment prove that Mv CCSC and KCCAMv SC have better performance than the classical Single-view K-means,and Multi-view K-means Concatenation(CK),and Multi-view K-means(MK),and Single-view Spectral Clustering(SC),and Spectral Clustering Concatenation(CSC),and Multi-view Spectral Clustering(MSC),which it has better clustering accuracy and robustness.