Image Sampling for Invariant Face Recognition

In this paper we present a novel scheme for reducing the impact of variations in head pose on face recognition. As its heavy reliance on the sampling center makes traditional log-polar sampling poorly suited for eliminating the impact of variation in head pose, we successfully overcome this problem using a quincunx pyramid sampling algorithm of our own design. Tests using the faces of 100 subjects from the color FERET database show that the algorithm we propose provides an accurate acceptance rate of 99.8% and a false acceptance rate of 0%.