Large Scale Double Density Dual Tree Complex Wavelet Transform Based Robust Feature Extraction for Face Recognition

Abstract

Varying effects in face recognition often causes intrapersonal variations due to which efficient feature extraction is desirable. In this work, a significant feature extraction technique based on Double Density Dual Tree Complex Wavelet Transform (DD_DTCWT) is proposed for face recognition. DD_DTCWT is a variant of wavelet transformation which provides better multiresolution sub-band spectral analysis of face images with good shift invariance and directional selectivity. Extensive experiments are performed with slight pose variations on ORL database and on images with illumination and expression variations in YALE database. It has been depicted from experimental results that the proposed technique for large-scale Double Density Dual Tree Complex Wavelet Transform-based robust feature extraction yields accurate results on Yale database and better results on ORL database as compared with state-of-the-art techniques. The classification is performed on training and testing face vectors based on DD_DTCWT using K-nearest neighbor classifier. Several experiments are performed to analyze significant features by varying decomposition levels for sub-band selection and number of images in training set.