DWT-based face recognition using fast walsh hadamard transform and chiral image superimposition as pre-processing techniques

One of the major challenges encountered by current Face Recognition (FR) techniques lies in the difficulties of handling varying poses and illuminations. In this paper we propose three novel techniques, viz., Fast Walsh Hadamard Transform (FWHT), Chiral Image Superimposition (CIS) and Discrete Wavelet Transform (DWT), to improve the performance of a FR system. FWHT is used for illumination normalization, CIS combats pose variances and DWT along with its sub-bands LL and LH are used for efficient feature extraction.

A Binary Particle Swarm Optimization (BPSO) based feature selection is used to reduce the number of selected facial features for recognition. Individual stages of the FR system are examined and an attempt is made to improve each stage. Experimental results show the promising performance of the proposed techniques for face recognition on four benchmark face databases, namely, Color FERET, ORL, CMU PIE and Extended Yale B.