Image-Based 3D Face Modeling from Stereo Images

This paper presents an automatic and novel method to generate a realistic 3D face model from stereo images. Typically, an image-based 3D face modeling system is in need of human intervention in facial feature extraction stage. To remove this human intervention, we propose HT(Hue-Tint) skin color model for facial feature extraction. Based on the proposed chrominance model, we can detect facial region and extract facial feature positions. Subsequently, the facial features are adjusted by using edge information of the detected facial region along with the proportions of the face. Moreover, the proposed facial extraction method can effectively eliminate the epipolar constraints caused by using stereo vision approach. In order to produce a realistic 3D face model, we adopt RBF(Radial-Based Function) to deform the generic face model according to the detected facial feature points from stereo images. For deformation locality parameter of RBF is critical since it can have significant impact on the quality of deformation. Thus, we propose new parameter decision rule that is applicable to scattered data interpolation. It makes clusters of feature points to detect points under the influence of each width parameter. From the experiments, we can show the proposed approach efficiently detects facial feature points and produces a realistic 3D face model.