Dictionary Pair Learning with Block-Diagonal Structure for Image Classification

Abstract

Dictionary learning has played an important role in the success of sparse representation. Although several dictionary learning approaches have been developed for image classification, the dictionary pair learning, i.e., jointly learning a synthesis dictionary and an analysis dictionary, is still in its infant stage. In this paper, we proposed a novel model of dictionary pair learning with block-diagonal structure (DPL-BDS), in which a block-diagonal structure of coding coefficient matrix and a block-diagonal structure of analysis dictionary are enforced. With the block-diagonal structures, discrimination of synthesis dictionary representation, coding coefficients and analysis dictionary are introduced into the dictionary pair learning model. An iterative algorithm to efficiently solve the proposed DPL-BDS was presented in this paper. The experiments on face recognition, scene categorization, and action recognition clearly show the advantage of the proposed DPL-BDS.