Efficient wavelet based scale invariant feature transform for partial face recognition
Even though, innumerable approaches have been proposed for holistic face recognition, problems caused by occlusions received less attention in the literature. However, partial faces frequently appear in many real time situations. Facial occlusions (by sunglasses, hat/cap, scarf, and beard) can significantly deteriorate the performances of face recognition systems under unconstrained scenarios. In such situations, algorithms developed under holistic face, results in catastrophic performance. In this paper, we have proposed a scale and rotation invariant wavelet feature transform for partial face recognition. Partial faces at different orientations are considered here for experimentation. Biorthogonal wavelet basis (4.4) is employed for obtaining the Discrete Wavelet Transform of the images. The scale invariant feature transform (SIFT) is then applied on low-low (LL) and high-high (HH) subbands of the images. Results obtained with wavelet SIFT method is compared with SIFT and appearance based face recognition technique (PCA) over (Milborrow / University of Cape Town) MUCT database. Experimental studies with 100 subjects show that the proposed method improves recognition accuracy and reduces false acceptance rate (FAR) and false rejection rate (FRR).
This publication has 18 references indexed in Scilit: