Nose tip detection from 3D facial mesh data using a rotationally invariant local shape descriptor

We present a nose tip detection method using a novel 3D local shape descriptor called DF (Distance based Fourier) descriptor that is rotationally invariant. The DF descriptor allows us to control the degree of descriptiveness depending on the complexity of object shape. When combined with SVM (Support Vector Machine), the DF descriptor proves powerful for nose tip detection. The detection method also features prescreening of candidate points for the nose tip using a constraint of being a protuberant point and 3D Harris Corner detection. The preliminary result for nose tip detection shows a great promise towards the detection of other fiducial features such as the eyes and the mouth corners and finally recognition of 3D faces.