Building 3D facial feature models by stereo and estimating its pose

摘要

We propose a framework for facial image processing performed in 3D space. We do intensity restriction in order to make the edge detection of the facial feature easier. Consisting of a sparse depth map is constructed from stereo images using segment-based stereo algorithm. Within the 3D model, data vertex and data arc are defined as two types of geometrical features. By searching for arcs of a specific curvature range from the model, we can locate the candidates of the irises, eyelids and the mouth. After that, the positions of the two eyes and the mouth are determined by checking their geometrical relations. Furthermore, The pose of a face is determined according to the positions of the two eyes and the center of the mouth.