Self-Representation Discriminant Analysis Dictionary Learning for Pattern Classification

As an important branch of dictionary learning, discriminant analysis dictionary learning can quickly acquire analysis sparse codes, which is widely used in pattern classification. To utilize the local feature information of samples flexibly, we propose a self-representation discriminant analysis dictionary learning (SrDADL) method. Specifically, we first design a self-representation discriminant term that transmits the geometric feature information of samples to analysis dictionary, which ensures that similar samples have similar analysis sparse codes under the action of the analytic dictionary. Then, we design a discriminative sparse code error term that forces the analysis code coefficient matrix to realize a block diagonal structure. We simultaneously combine the two terms into the analysis dictionary learning model to improve the discrimination capability of the analysis dictionary and the coding coefficients. We also design an efficient iterative algorithm to solve the optimization problem of SrDADL. Extensive experiment data show that SrDADL is effective for pattern classification tasks.