A Improved Fuzzy K-Subspace Clustering and Its Application in Multiple Dominant Sparse Component Analysis

Sparse Component Analysis (SCA) has been successfully applied in Blind Signal Separation (BSS). The two-stage approach is proved to be useful in dealing with SCA problem. In order to solve some common problems in the mixing matrix estimation stage, a novel algorithm is proposed in this paper, which is not only able to implement subspace clustering but also capable of detecting the number of hidden subspaces. Extensive computer simulations demonstrate the efficacy of the proposed method.