3D Facial Feature Location with Spin Images

This paper shows an original 3D facial feature localization method. The localization of feature points in the face is a relevant step that allows the normalization of the data in size and orientation. This stage is essential previous to a range data calculation or a verification process. The localization procedure proposed is based on both, clustering techniques over discrete curvatures calculated, and Spin Images as a global registration method. This method has been tested with a 3D Face Database acquired by a laser scanner in several conditions of position and gesture. The aim is to find three specific feature points: nose-tip and left eye and right eye inside corners. Results show a success rate of 100% and a low computation time allowing the system to work in real time.