Block recursive least squares dictionary learning algorithm

The block recursive least square (BRLS) dictionary learning algorithm that dealing with training data arranged in block is proposed in this paper. BRLS can be used to update overcomplete dictionary for sparse signal representation. Different from traditional recursive least square algorithms, BRLS is designed for data in a block form and the recursion is developed without using the matrix inversion lemma. The proposed algorithm is applied in synthetic data and real image reconstruction. Simulation results show that the new algorithm achieves a better performance than traditional approaches.