A compact local binary pattern using maximization of mutual information for face analysis

Although many variants of local binary patterns (LBP) are widely used for face analysis due to their satisfactory classification performance, they have not yet been proven compact. We propose an effective code selection method that obtain a compact LBP (CLBP) using the maximization of mutual information (MMI) between features and class labels. The derived CLBP is effective because it provides better classification performance with smaller number of codes. We demonstrate the effectiveness of the proposed CLBP by several experiments of face recognition and facial expression recognition. Our experimental results show that the CLBP outperforms other LBP variants such as LBP, ULBP, and MCT in terms of smaller number of codes and better recognition performance.