模式识别与人工智能.doc

基于曲面等高线特征的不同姿态三维人脸深度图识别 中图分类号：TP391 文献标识码 A 3D Facial Depth Map Recognition in Different Poses with Surface Contour YE Chang-ming1 JIANG Jian-guo1,2 ZHAN Shu1,2 SHIGERU ANDO3 1 （ School of Computer Information, Hefei University of Technology, Anhui Hefei 230009, China） 2 （Engineering Research Center of Safety Critical Industrial Measurement and Control Technology, Ministry of Education, Anhui Hefei 230009, China） 3 （School of Information Science, the University of Tokyo, Japan Tokyo） Abstract: Traditional face recognition methods are often based on the information of each pixel in human facial images，and due to the large amounts of image information, the dimensionality reduction is necessary. However, these dimensionality reduction methods are always abstract and meanwhile time cost is considerable. This paper presents a novel face recognition method —FDAC, which is base on correction “mesion” depth maps from the 3D facial imaging system in different poses. Firstly, depth maps are aligned under differential geometry theory guidance, and the human face is described by the contours. Secondly, Fourier descriptor assists to extract the facial features. Finally, these features from contours are used in face recognition process. Experimental results show that FDAC performs well in recognition accuracy and especially reduce much time cost in compare with Eigenface method. Key words：Fourier descriptor; face recognition OCIS codes: 引言 人脸识别特指利用分析比较人脸视觉特征信息进行身份鉴别的计算机技术。静脉识别基于特征脸（PCA）的人脸识别方法电子商务社会福利保障安全防务等领域。——FDAC（Fourier Discriptor and Contour）方法。该方法基于实时人脸成像系统获得的三维人脸深度图，在对不同姿态人脸深度图进行校正的基础上，使用等高线和傅里叶描绘子提取三维人脸图像的脸部特征，并将这些特征用于人脸识别。实验结果表明，与传统的人脸识别方法相比，新方法在识别率和时间开销上均有较好的表现。 不同姿态下人脸深度图的校正 本文使用的人脸深度数据均通关过三维人脸实时成像系统获得。改成像系统基于相关型图像传感器（CIS） 图1 三维人脸实时成像实验系统实物图 Fig1. The experiment system of 3D real-time facial imaging. 图2 三维人脸图像:灰度图像 左 和深度图像 右 Fig2. 3D facial data: Intensity data and Depth data 在实际的操作中人脸往往并非正对摄像头，采集得到的深度图往往在空间的各个方向存在偏转或者移动，并不是标准的“正中面”图像。为了获得更好的识别效果，需要利用三维