부산대학교 도서관

An effective face descriptor is critical for a successful face recognition system and must overcome the challenges of changing environment. The face representation must have discriminatory information and be computationally feasible for any face recognition system. In this paper we propose a new face descriptor, Similarity Pattern of Image Directional Edge Response (SPIDER), for face recognition. An image is divided into smaller local regions and 8 directional edge responses are generated for each pixel position in the regions. The regional cumulative response of each direction is calculated and a histogram is generated consisting of 8 bins, one for each of the directions. The SPIDER code is generated by calculating the similarity between the histogram of the local region around each pixel against the histogram of neighbor regions. The feature vector is projected to a low-dimension vector space using a dimension reduction method to minimize the classification time. Experiments using the proposed method were carried out on the FERET database and results show improved recognition rates indicating the robustness to changing environment, and a low classification time compared to the existing methods.