Fusion of Moving and Static Facial Features for Robust Face Recognition

This paper presents a robust and efficient face recognition technique. It consists of six major steps. Initially the proposed techniques extract salient facial landmarks (eyes, mouth, and nose) automatically from each face image. SIFT descriptor is used to determine all facial keypoints from each landmark. These feature points are considered to obtain a relaxation graph. Given two relaxations graphs from a pair of faces, matching scores between two corresponding feature points has been determined with the help of iterative graph relaxation cycles. Dempster-Shafer decision theory is used to fuse all these matching scores for taking the final decision. The proposed technique has been tested against the three databases, namely, FERET, ORL and IITK face databases. The experimental results exhibit robustness of the proposed face recognition system.