Robust SIFT for dark face images recogntition
Meriama Mahamdioua, Mohammed Benmohammed · 2016
Scale Invariant Feature Transform (SIFT) method is used to detect and describe local images features (keypoints) that are invariant to scale, rotation, translation and partially invariant to image illumination changing. However, this method gives unsatisfactory results under deteriorated lighting conditions. In other hand, the basic matching method of SIFT can produce false matched features, which leads to false matched objects. To improve the performance of SIFT in this situation, we propose in this paper to use a preprocessing method based on Gaussian filter and an amelioration of TT [5] to eliminate the variation of illumination, and a modified matching method to remove the false keypoints matched. Our proposed method is compared with a set of illumination normalization techniques (SSR, WD, SSQ, HOMO, TT and MSW) applied on dark face images. The experiments results confirm the superiority of the proposed method compared with the tested ones for face recognition under uncontrolled lighting conditions.