Design of Discriminative Dictionaries for Image Classification

Abstract

In recent years, sparse representation theory has been widely used in the field of image classification. Based on kernel trick and the improved Fisher discrimination dictionary learning, this paper designs a dictionary learning algorithm that can effectively improve image classification performance. Kernel space transformation can learn non-linear structural information, which is very useful for image classification. The traditional kernel dictionary learning algorithm has high computational complexity, which is not conducive to practical application. We address this problem by proposing a sample preprocessing method based on Nystrom algorithm. By introducing the incoherent promoting terms into the Fisher discrimination dictionary learning model, we can obtain more discriminative coding coefficients while learning a structured dictionary. The effectiveness of the proposed kernel incoherent Fisher discrimination dictionary learning (KIFDDL) method is verified by the results of the classification experiments on several publicly image databases.