A Truncate-FCM Algorithm for Dictionary Generation

Abstract

Learning over complete dictionaries for sparse signal/ image representation has become an extremely active area of research in the last few years. In this paper, we present a novel method involving an iterative process that alternates between a cluster step solved by Fuzzy C-Means clustering (FCM) algorithm and a truncate step for the weight coefficients of each cluster. It benefits from the adaptability to the training signal samples through clustering and takes advantage of the sparsity by a truncate operation. Numerical experiment in image denoising shows that the proposed algorithm is comparable to the K-SVD, which is a well-known dictionary design or generation method.