Locating facial feature points using support vector machines

Recent emergent face-related applications, such as face recognition and face animation, normally demand robust and accurate extraction of facial feature points as a preliminary step. In this paper, we propose an accurate facial feature point extraction algorithm that combines the good generalization ability of support vector machine (SVM) as well as the binarization techniques with color criterion. In the proposed algorithm, we first utilize the SVM detector to find the regions that contain the desired features such as eyes and mouths. Then, by properly binarizing these feature patterns and using the color information in the detected local regions, the facial feature points can be successfully located. Images from CMUPIE face database together with four kernel types of SVM and three different types of image representation are tested in the experiments for comparison. Experimental results demonstrate the performance of the proposed algorithm.