一种面向三维人脸识别的同一截面有效轮廓线提取算法

摘要

In light of the study on the robustness of 3D face recognition system for changes in expression and guise,we propose a face recognition method which is based on effective contours on same section of faces.First,the method corrects the face roughly based on the normal vector of the nose tip area manually calibrated,and then makes precise guise correction based on the ICP algorithm of the front faces with natural expression in same standard.Next,the method calibrates the precise position of nose tip,and extracts two contour lines of horizontal and vertical each passing the nose tips on the same section of different faces.The threshold method is used to extract the effective contours passing the nose tips,and the ICP algorithm is used to calculate the similarities,the recognition results of two contours are then fused.Experimental results show that on CASIA 3D face database the method has better robustness on the changes in expression and guise.%针对三维人脸识别对表情及姿态变化的鲁棒性研究，提出一种基于人脸同一截面有效轮廓线的人脸识别方法。首先根据手工标定鼻尖点区域的法向量对人脸进行粗略矫正，再基于同一标准正中面人脸的迭代最近点（ICP）算法进行精确姿态矫正，标定精确鼻尖位置，提取过鼻尖的不同人脸同一截面横纵两条轮廓线。用阈值法提取过鼻尖点的有效轮廓线，采用 ICP 算法计算相似度，对两条轮廓线识别结果进行融合。实验结果表明，在 CASIA 3D 人脸库上对表情及姿态变化有较好的鲁棒性。