Joint local constraint and fisher discrimination based dictionary learning for image classification

Dictionary learning has achieved outstanding results in the field of pattern recognition. The locality as an important structure characteristic of the data has been widely used in various learning task including dictionary learning, and Fisher discriminant has also been widely used. In this paper, to improve the performance of dictionary learning, we propose a joint local-constraint and fisher discrimination based dictionary learning method (JLCFDDL) for image classification. Our method uses the Laplacian regularized constraint of atoms rather than that of training samples to preserve the local information of the data. Meanwhile, fisher discriminative constraint is imposed on the atoms to maintain the differences between the atoms of different classes and to reduce the differences of atoms of the same class. The joint constraints make the obtained dictionary with powerful image classification performance. We also provide mathematical analysis of the proposed objective function. A large number of experiments prove that our method achieves better performance than existing state-of-the-art dictionary learning method. (C) 2019 Elsevier B.V. All rights reserved.