Face recognition using adaptive filter wavelet transform based feature extraction

Face Recognition (FR) under varying pose, illumination and occlusion conditions is challenging. In this paper, a novel algorithm called Mirrored Fusion is proposed to normalize the effects of pose variations in facial images. A unique feature extraction technique called Adaptive Filter Wavelet Transform (AFWT) is proposed, which is a combination of Stationary Wavelet Transform (SWT) along with Wiener filtering and scaling, Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT). AFWT results in low contrast images with prominent features which are desirable for enhanced recognition rate of the FR system. Experimental results obtained by applying the proposed algorithm on FERET, Pointing Head Pose, CMU PIE, Extended YaleB and Georgia Tech face databases show that the proposed system outperforms other FR systems.