Discriminative local difference patterns for robust face recognition

摘要

A novel facial feature descriptor, termed discriminative local difference patterns (DLDP), is proposed for robust face recognition. DLDP extracts local difference patterns (LDP) based on directional texture operators. To improve the compactness and discriminability of LDP, a two-stage linear discriminant analysis for dimensionality reduction is further applied. The proposed DLDP effectively captures the local and holistic characteristics of the face image and shows great tolerance to illumination changes, robustness to variations in pose and facial expression, and computational efficiency. Experimental results on challenging face databases show that DLDP consistently outperforms current state-of-the-art facial feature descriptors.