Face recognition robust to head pose changes based on the RGB-D sensor

Face recognition is still a great challenge in biometrics research, because of the large variations of facial appearance, caused by head pose, lighting, facial expression, aging, etc. Among all possible variations, the biggest change of facial appearance in 2-dimensional (2D) face images probably comes from the three-dimensional (3D) head rotations. With the sensor technology advances, e.g., the recent RGB-D cameras, we study the advantage of using RGB-D images for face recognition, focusing on the challenge of 3D head pose variations. We propose an approach to face recognition robust to head rotations utilizing the RGB-D face images. Unlike the traditional 3D morphable model, our method does not need to learn a generic face model or take a complicated 3D to 2D face registration. We study what is the appropriate scheme to deal with pose variations in order to develop a robust system towards pose-invariant face recognition. Experiments on a public database show that our approach is effective and efficient for face recognition under significant pose changes. Our preliminary result demonstrates the advantages of using the RGB-D sensor for face recognition robust to large pose variations.