Double δ-LBP: A novel feature extraction method for facial expression recognition

Abstract

The Local Binary Pattern (LBP) is a widely used descriptor in facial expression recognition due to its efficiency and effectiveness. However, existing facial expression recognition methods based on LBP either ignore different kinds of information, such as details and the contour of faces, or rely on the division of face images, such as dividing the face image into blocks or letting the block centering on landmarks. Considering this problem, to make full use of both detail and contour face information in facial expression recognition, we propose a novel feature extraction method based on double δ-LBP (Dδ-LBP) in this paper. In this method, two δ-LBPs are employed to represent details and the contour of faces separately, which take different kinds of information of facial expression into account. Experiments conducted on both lab-controlled and wild environment databases show that Dδ-LBP outperforms the original LBP method.