基于正方形切平面描述符的三维人脸区域标记

3D Face Labeling Based on Square Tangent Plane Descriptor

针对目前三维人脸模型的语义标记和分割研究较少的问题,提出一种基于正方形切平面描述符的三维人脸模型区域标记算法。这种新的描述符由三维人脸模型顶点的正方形切平面区域内的几何信息编码而成。随后通过随机森林算法对其进行学习,对模型上的所有顶点进行分类,从而实现对三维人脸模型上眉毛、眼睛、鼻子、嘴巴等区域的识别和标记。在定位仿真实验中,分类准确率可达94.46%。该描述符具有旋转、头部姿势与三维模型分辨率不变性,对模型噪声具有鲁棒性。实验结果表明,该方法能有效标记三维人脸模型区域。

Aiming at the shortcomings of current research on semantic marking and segmentation of 3D face models, a 3D region labeling algorithm based on square tangent plane descriptors was proposed. This new descriptor was obtained by encoding the geometric information in the square area on the surface of 3D face mesh model. Then it was learnt by the ran- dom forest algorithm to realize the classification of the vertex on 3D face model, so as to identify and mark the eyebrows, eyes, nose, mouth and other regions. Simulation result can achieve a classification accuracy of 94.46 %. The proposed de- scriptor has rotation, head pose, 3D model resolution invariance and is robust to noise. Experimental results show that the proposed method can effectively mark the 3D face model region.