A contourlet-transform (CT) based sparse independent component analysis for blind image separation is proposed. The images are first decomposed into sets of local features with various degrees of sparsity, and then the intrinsic property is used to select the best (sparsest) subsets of features for further separation. Based on sparse description of the contourlet-transform, the proposed approach is able to yield better performance, including faster convergence and the certain order for the separated signals. Simulation results confirm the validity of the proposed method.
展开▼
