Human facial feature localisation by Gabor filter and clustering

Human facial features localization is an important process of face recognition, since it helps generating face images in accordance with specified criteria, or building unique face model. This paper presents a novel method for finding facial features through Gabor filtering and k-means clustering analysis. By Gabor filtering, face images are transformed into magnitude responses. In magnitude responses, areas containing facial features demonstrate relatively strong responses. After thresholding magnitude responses, strong responses are remained, but weak responses are neglected. Points belonging to facial features are collected for the k-means clustering. Points are grouped into different clusters. Each cluster corresponds to a facial feature. By testing on the ORL face database, the method shows its accuracy and rapidness on locating facial features, such as eyes, nose, and mouth. It also displays its robustness on people who have thick beard or moustache.