Automatic face recognition is one of the most active and challenging research topics in pattern recognition and computer vision, which has gained much attention during the past few years. Most of proposed 2D based face recognition systems are designed to work with frontal face images. Since human face has a 3D nature, all of information about the structure of the face is not present in the frontal view. Facial profile curve contains additional depth information of the face that is not presentin the frontal view. In this paper we present a novel and robust method for human identity recognition from the face profile view. The proposed method uses the center of ear and the tip of noise as the two reference points. The face profile curve is then extracted by a segmentation method based on 2D histogram of the individual's face skin in Cb and Cr color channels. Having the two reference points, an angular sampling method is used to extract final normalized feature vector from the facial profile curve points. In the matching phase the Hausdorff distance metric makes the algorithm robust to small displacements in reference points detection. The performance of this method illustrated using CBL database.
