Invariant Facial Features Under Pose Variations for Face Recognition

Pose variation is one of the major challenges in face recognition. In this paper, two global invariant facial features are proposed: (1) horizontal facial feature; (2) vertical facial feature. The paper proves that the proposed two facial features are invariant under pose variations. Cosine-face, extracted by performing a method based on a combination of discrete cosine transform (DCT) and principal component analysis (PCA), is used as the basic feature in face recognition application. By integrating proposed invariant features and cosine-face, unified facial features are obtained to represent each face image. Experimental results on Cambridge ORL face database show that substantial improvements are obtained by using our proposed global invariant features