Semi-supervised Sparse Subspace Clustering Based on Re-weighting.

The traditional sparse subspace clustering algorithms are easily affected by the similarity matrix, which may lead to different clustering results by different similarity matrix. That is to say, constructing a reasonable similarity matrix is the key to sparse subspace clustering. Based on reweighting subspace clustering, a semi-supervised sparse subspace clustering algorithm based on re-weighting is proposed in the paper. Firstly, the global similarity structure of the data can be better captured by constraining the coefficient between the cannot-link labels to be 0. Secondly, with the help of reweighted l1-norm minimization sparse optimization framework, the adaptive similarity matrix can be obtained. Furthermore, the above similarity matrix can be adjusted by using prior information. Experimental results indicate that the proposed clustering algorithm is more efficient than other clustering algorithms on benchmark data sets. [ABSTRACT FROM AUTHOR]