Current Proceedings on Technology

Yazarlar: Hasan Serhan Yavuz

Konular:-

Anahtar Kelimeler:Face recognition,Face template,Image transformation,Appearence based recognition methods

Özet: Face recognition is one of the most popular pattern recognition research subjects which can be defined as the identification of individuals from a set of face images. Appearance-based face recognition methods represent the face by using several raw face images usually in terms of gray level intensity values. In face recognition applications, most of the researchers prefer to use standard databases in which the face regions have already been cropped. In this study, we concentrate on preparation of face templates to be used with the traditional subspace based recognition methods. The templates are prepared from the portrait images of 100 people photographed in our lab including 18 frontal poses per person taken under different illumination conditions and mimic variations. We used some image transformation techniques such as shifting, rotation and scaling to map the face region of the portrait image into previously designed three different face templates by using the locations of manually annotated fiducial points. After that, we performed face recognition experiments by using Eigenfaces, Fisherfaces and the discriminative common vectors methods which are among the most popular subspace techniques in the area. At the end, we point out how different face alignment procedures effect the recognition rates and state the most suitable template style to be used with the appearance based recognition methods.

---