Abstract
In this paper the problem of sparse source separation of linear mixtures is addressed. We propose to apply K-SVD, which is a leading dictionary learning method, for this purpose. Further, a modified gradient-based K-SVD scheme for incoherent dictionary learning and source separation is proposed. The promising results on random synthetic signals reveal the ability of this technique for utilizing in source separation framework. We also suggest BOLD detection fMRI as an application for this method. The preliminary results confirm the successful separation of this type of data.