A collaborative representation face classification on separable adaptive directional wavelet transform based completed local binary pattern features

Face recognition has emerged as the most active area of research in computer vision. A variety of face recognition methods were devised, though several challenges are imposed due to face variations such as facial expression, pose variation and illumination variation which generate great concern in developing efficient face recognition methods. It is desirable to extract robust local descriptive features to effectively represent such face variations. The essential attribute of the proposed method is to extract directional descriptive local features based on the face image characteristics. In order to extract the multi-resolution directional features as per the face variations, a 2-D interpolation-based separable adaptive directional wavelet transform (SADIWT) is proposed. For the implementation of 2-D SADIWT, a set of seven directions with an improved quadtree partitioning scheme is proposed. Completed local binary patterns (CLBP) superior to local binary patterns (LBP) in extracting local texture features are applied on top level's 2-D SADIWT sub-bands to obtain local descriptive features. Collaborative representation classification (CRC) takes benefit of these descriptive features and leads to a very competitive classification performance. Extensive experimental results on benchmark face databases such as ORL, FERET, CMU-PIE, and LFW demonstrate high classification accuracy of the proposed method. A comparison with numerous methods which include various holistic, LBP-based descriptors and representation methods demonstrate the efficacy of the proposed method. Experiments are also conducted to exhibit the robustness and discrimination capability of the proposed method for handling single image per person (SIPP) and random block occlusion problem.