Fusion of Zernike Moments and SIFT Features for Improved Face Recognition

Combining the feature sets that are invariant to global as well as to local variations of face images would be an efficient approach to construct an optimal face recognition system. Thus, identification and combination of complementary feature sets has become an active topic of research in recent days. In this paper, a combination of two useful methods, i.e. Zernike Moments (ZMs) and Scale Invariant Feature Transform (SIFT) has been proposed for the recognition of face images wherein the global information of face images has been effectively extracted by the ZMs approach while SIFT descriptor is used to locate local distinct keypoints. Exhaustive experiments are performed on ORL and Yale face databases. It has been observed that the proposed fusion achieves 98.5% and 91.67% recognition rates on ORL and Yale databases respectively. The inherent characteristics of ZMs and SIFT are retained in the combined descriptor and therefore the proposed approach is highly robust against pose, illumination and expression variations.