Joint Collaborative Representation and Discriminative Projection for Pattern Classification

Representation-based classifiers have shown the impressive results for pattern classification. In this paper, we propose a joint collaborative representation and discriminative projection model (JCRDP) for subspace learning. We aim to seek a linear projection matrix to effectively reveal or maintain the underlying structure of original data and well fit collaborative representation classifier simultaneously. Unlike previous representation-based subspace learning methods, in which the linear reconstruction and the generalized eigenvalue decomposition are two independent steps, our proposed JCRDP integrates these two tasks into one single optimization step to learn a more discriminative linear projection matrix. To effectively solve JCRDP, we develop an alternative strategy to deal with the optimization problem. Extensive experimental results demonstrate the effectiveness of our proposed method.