IBP-SVD: A practical method for learning adaptive dictionaries for image de-noising

Abstract

In recent years, there is a growing interest in the research of sparse representations for signals over an overcomplete dictionary. The Dictionaries can be either pre-specified transforms or designed by learning from a set of training signals. The K-SVD is a dictionary training algorithm recently proposed. However, It can not find the truly sparse representations in sparse coding stage. We analyze the relationship between matching pursuit and basis pursuit algorithms and present another practical method, called IBP-SVD, which can find the sparsest representations frequently. It effectively improves the training speed and the precision of the trained dictionary. Experimental results of image de-noising show that IBP-SVD has a better performance than K-SVD method and reduces time in the process of learning dictionaries.