Adaptation of SIFT features for face recognition under varying illumination

Scale Invariant Feature Transform (or SIFT) is an algorithm used to detect and describe local features in images invariant to image scale, translation and rotation. All SIFT-based face recognition techniques found in literature so far rely heavily on the keypoint detector. The purpose of this detector is to locate interest points in the given image that are later used to compute the SIFT descriptors. While these descriptors are known to be among others (partially) invariant to illumination changes, the keypoint detector is not. To overcome the presented shortcoming of SIFT-based methods, a novel face recognition technique is proposed in this paper. The SIFT descriptors are computed at fixed points in the locations of the nodes on a regular grid, overlapping face image. By doing so, the need for keypoint detection on the test images is eliminated and greater robustness to illumination variations is achieved in comparison with related approaches from the literature. Experimental results, obtained on the Extended Yale face database B, demonstrate that better results are achieved with proposed technique in comparison with the remaining techniques assessed in our experiments, especially under severe illumination conditions.