Improved face recognition system using probability distribution functions extracted from wavelet subbands

In this paper, we propose a high performance face recognition system based on the probability distribution functions (PDF) extracted from discrete wavelet transform in different colour channels. The PDFs of the equalized and segmented faces in different subband images obtained from discrete wavelet transform (DWT) are used as statistical feature vectors for the recognition of faces by minimizing the Kull-back- Leibler divergence (KLD) between the PDF of a given face and the PDFs of faces in the database. Majority voting (MV) and feature vector fusion (FVF) methods have been employed to improve the recognition performance by combining feature vectors in HSI and YCbCr colour spaces of LL, LH, HL, and HH subband images. The system has been tested on the FERET and the head pose (HP) face databases. The results have been compared with conventional PCA, and three state-of-art face recognition techniques, namely, adaptive local binary pattern (LBP) PDF based face recognition, nonnegative matrix factorization (NMF), and supervised incremental NMF (INMF).