Lucas-Kanade Scale Invariant Feature Transform for Uncontrolled Viewpoint Face Recognition

Face recognition has been widely investigated in the last decade. However, real world application for face recognition is still a challenge. Most of these face recognition algorithms are under controlled settings, such as limited viewpoint and illumination changes. In this paper, we focus on face recognition which tolerates large viewpoint change. A novel framework named Lucas-Kanade Scale Invariant Feature Transform (LK-SIFT) is proposed. LK-SIFT is an extension of SIFT algorithm. SIFT is a scale and rotation invariant algorithm, which is powerful for small viewpoint changes in face recognition, but it fails when large viewpoint change exists. To handle this problem, we propose to use Lucas-Kanade algorithm to generate different viewpoint face from a single frontal face. After that, SIFT is used to detect local features from these viewpoints, these SIFT features contain information of different viewpoint face, which can deal with the problem of face viewpoint change. Finally, our framework is compared with the SIFT algorithm and other similar solutions. Experiment results show our framework achieves better recognition accuracy than SIFT algorithm at the cost of acceptable computational time gains compared with other similar algorithms.