A Dictionary Learning Method for Sparse Representation Using a Homotopy Approach

Abstract

In this paper, we address the problem of dictionary learning for sparse representation. Considering the regularized form of the dictionary learning problem, we propose a method based on a homotopy approach, in which the regularization parameter is overall decreased along iterations. We estimate the value of the regularization parameter adaptively at each iteration based on the current value of the dictionary and the sparse coefficients, such that it preserves both sparse coefficients and dictionary optimality conditions. This value is, then, gradually decreased for the next iteration to follow a homotopy method. The results show that our method has faster implementation compared to recent dictionary learning methods, while overall it outperforms the other methods in recovering the dictionaries.