Facial information retrieval using component-based classification and scale invariance

There are many applications in which identification by facial images is desired. Because of the variations in faces as well as the images that capture this human feature, recognition and identification of facial images is a complex problem. In this paper, a face identification and recognition system is designed using scale invariant points and phase extraction. Identification of a face within an image is the first step; here a skin color model is used by matching the texture of skin via hue and saturation values. This step is followed by a face recognition operation whereby components of the face (e.g., eyes, lips, and nose) are extracted using a geometric approach rather than holistic approach. Each individual component of the face is extracted using separate classifiers resulting in individual feature signatures. The approach of extracting signatures for each facial component results in classifying scale invariant images of the face which provides robust recognition among clusters and occlusions in real-time. The output of each classifier is individually processed for features that are invariant of scale, translation, rotation, and partially invariant to illumination changes. Results show that the constructed PID's (Phase IDentifications) are unique for each component of a face and the signatures of a facial image obtained via this hybrid face recognition approach provides better results than classical methods by focusing on specific areas that are sensitive to variations. The main advantage of PIDs is that they are applicable to a broader range of image structures, such as corners and edges, for which scale selection is unreliable in current scale invariance algorithms. In this paper, we extend our work in PIDs by focusing on the selection of the number of key points in the facial image in order to test robustness and accuracy. This is demonstrated by several face detection examples.