Face Detection And Tracking In Color Image Sequences

Automatic face detection and tracking is an important and front-line subject in computer vision domain. In this paper, we have made a deep study on automatic face detection and tracking in color image sequences. A systematic summary of previous work has been given first. Then this paper presents a novel multi-stage face detection algorithm, which makes a good use of human face pattern's valuable information in colour image sequences. The difficult detection task has been divided into four steps: the preprocessing, which is to gain skin colored regions with human skin color model; the roughly detection and face region refining by elliptic curve fitting; the fine detection with facial features' detection and location; the face/non-face classification step based on PCA and Gaussian density estimation technique. Experimental results demonstrate the detection algorithm is efficient in complex background. Finally, this paper also presents a new method for face tracking. The method is novel since it combines motion estimation technique and adaptive template matching technique, and the above two techniques are both based on Kalman filter. This face tracking system enables the robust tracking of face with translation, rotation, zoom in/out, and has a special attention to occlusion handing.