Expression invariant face recognition using local binary patterns and contourlet transform

Face recognition is one of the most widely used biometric techniques in surveillance and security. Although many state-of-the-art systems have already been deployed across the world, the main challenge for feature extraction comes from the variations in the query images captured under uncontrolled situations. Unlike the local binary patterns or steerable pyramids which construct a feature vector strictly from spatial and transform domain, respectively, our approach built a method that exploits the features from both spatial as well as contourlet transform domain. Specifically, the contourlet transform exhibits properties like directionality and anisotropy and hence, results in extraction of significant features. Furthermore, we have proposed a novel coefficient enhancement algorithm which is applicable on the contourlet subbands to make the system more robust by enhancing skin region features. In addition, we show that the feature level fusion produces a robust feature vector, which yields competitive face recognition rates on the Cohn–Kanade (CK), Yale, JAFFE, ORL, CMU-AMP and our own face database. Finally, we benchmark our approach with other contemporary approaches and found it as most robust expression invariant face recognition technique.