A novel multi-pose face recognition via robust SIFT feature
Xinao-Bing Xian, Hua-Juan Wu, Mingxin Zhang, Jin-Long Zhang, Xuedan Zhan · 2013
The performance of face recognition algorithm significantly degrades when the pose of probe face is different from gallery face, especially when the angular difference between them is larger than 45°. One of the possible solutions is that not only using frontal face but combining frontal and profile face images as gallery images. According to this idea, this paper proposes a simple, efficient robust SIFT feature method, which generates the face feature database (FFD) with multi-pose face images. The feature vectors are extracted from multiple poses of each person's face by using SIFT algorithm. Then, by computing the dot product of each feature vector with all others, the robust features which constitute the FFD could be identified. Meanwhile, in the proposed scheme, the importance of features is considered by assigning different weights, which improves accuracy. Experimental results on the PEI and the CMU PIE database demonstrate the effectiveness of the proposed method.