Face pose estimation using distance transform and normalized cross-correlation

Face pose estimation plays a vital role in human-computer interaction, automatic human behavior analysis, pose-independent face recognition, gaze estimation, virtual reality applications etc. A novel face pose estimation method using distance transform and normalized cross-correlation, is presented in this paper. The use of distance transform has two main advantages: first, unlike intensity image, distance transform is relatively invariant to intensity and illumination of image. Secondly, unlike edge-map the distance transform has a smoother distribution. The distance transform property of being invariant to intensity makes the proposed method suitable for different skin colors. Since distance transform is relatively invariant to illumination, the proposed method is also suitable for different illumination conditions. Further advantages are that the proposed method is relatively invariant to facial hairs and whether the subject is wearing glasses. The proposed method has been tested using the CAS-PEAL pose database with very good results.