一种基于多视点图象的可变姿态人脸识别系统

本文报告了一种多姿态人脸图象识别原型系统，它不同于现有系统和方法，该系统可工作在合作对象下允许姿态变化（存在图象平面内旋转和深度方向上旋转，限于双眼可见）的人脸图象识别。由于对成象条件有所放松，故可望应用于身份验证、保安和视频会议等领域。对姿态可变条件下的人脸特征检测、姿态估计、识别建模以及基于模板相关的匹配等技术进行了深人研究，分析了光照、姿态及分辨率变化等因素对识别的影响程度。实验结果表明，对于３０类人脸，每人１８幅图象大小的测试集，达到了１００％的识别率。

A View-based Automatic Face Recognition System

n this paper, an automatic human face recognition system is described. Unlike most of the existing systems that can only operate on the frontal view of face or the facial profile,our system can workunder varying pose with less rigid conditions. Basically,the face recognition system is based on multi-view template matching technique. In order to tackle the problems of absence invariant to malization, by which the confounding effects are first isolated, and then the faces are normalized prior to the Independent classification. The performance of our system to recognize faces with varying postures has been demonstrated by operating on a database of 30 individuals, 18 images for each person, with 100% recognition ratio.