Paper : Person-Specific SIFT Features For Face Recognition

Scale Invariant Feature Transform(SIFT) is widely and successfully applied to object detection and recognition. However, the representation ability of SIFT features in face recognition has rarely been investigated. This paper proposed to use the person-specific SIFT features and a simple matching strategy combine with local and global similarity on key-points cluster to solve face recognition.

The kernel of this paper is to cluster the key-points extract by SIFT, using a cluster algorithm: K-mean, to construct stable effective sub-regions on face image. Modify from the original SIFT matching, this paper use a combination of Local and Global similarity to match each sub-region with the face DB.

The experiment shows that this method perform well in expression variations, but failed in illumination variations, because the person-specific features may be sensitive then. In other hand, the method shows well performance in large pose view angle, and this is the most contribution of this paper.