Face identity verification based on sinusoidal projection

Abstract

This paper proposes a technique for face feature extraction using sinusoidal projection. Essentially, the technique uses a projection matrix, which is formed by stacking vectors with sinusoidal values at different frequencies, to directly multiply with raw image matrix for weighted feature extraction. Orthogonality among vectors within the sinusoidal projection matrix is observed when the frequencies are chosen as multiples of the fundamental frequency. The proposed technique shows promising verification performance on three face databases.