DCT pyramid based face recognition system

Face recognition (FR) is affected by various factors such as change in illumination, pose, expression, aging and various backgrounds. In this paper face recognition system based on DCT pyramid feature extraction is presented. We applied DCT pyramid on each face image to decompose it into approximation subband and reversed L-shape blocks. Then simple set of block based statistical measures were calculated on various subbands obtained by DCT pyramid. This simple set of statistical measures is efficient way to reduce dimensionality of the feature vector and to improve performance of classification. Experimental results on standard FERET database shows that proposed system achieves more accurate results than variable block based statistical measures. Moreover, it outperforms with the well known method such as block based DCT with zigzag scanning in terms of accuracy and recognition rate.