基于正方形描述符和LSSVM的三维人脸区域标记

3D face region labeling based on square descriptor and LSSVM

提出了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。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。通过最小二乘支持向量机算法(LSSVM)对其进行学习,对三维模型上所有顶点进行分类,实现了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。定位仿真实验中,分类准确率可达92.35%。所提描述符具有旋转、头部姿势、三维模型分辨率不变性,对模型的噪声具有鲁棒性。实验结果表明:提出的方法能有效标记三维人脸模型区域。

A 3D face region labeling algorithm based on square tangent plane descriptors is proposed. This new descriptor is obtained by encoding the geometric information in the square area on the surface of 3D face mesh model. It is learnt by the least squares support vector machine （LSSVM） 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 92.35 %. The proposed descriptor has rotation, head pose,3D model resolution invarianee and it has robustness to noise. Experimental results show that the proposed method can effectively mark the 3D face model region.