基于仿射模板匹配的多角度单人脸定位  

Single Rotated Face Location Based on Affine Template Matching

作　　者：梁路宏[1,2] 艾海舟[1,2] 何克忠[1,2] 张钹[1,2] 

机构地区：[1]清华大学计算机科学与技术系,北京100084 [2]清华大学智能技术与系统国家重点实验室,北京100084

出　　处：《计算机学报》2000年第6期640-645,共6页Chinese Journal of Computers

基　　金：国家重点预研基金!( 7B8);国家"八六三"高技术研究发展计划!( 863 -5 12 80 5 110 0 )

摘　　要：提出了一种基于仿射模板匹配的多角度单人脸定位算法 .用平均脸作为基模板 ,经由不同的伸缩比例和旋转角度构成的仿射变换生成各种姿态的模板 ,并将模板分为双眼模板和人脸模板对 ,首先使用双眼模板搜索符合眼睛的区域 ,再用整个人脸模板匹配进一步检验是否符合人脸模式 ,并利用马赛克规则验证 ,给出最佳匹配作为人脸位置 .对包含 144幅图像和 6 0幅图像的两个图像库的定位准确率分别达到 79.2 %和 85 % ,表明该算法的有效性 .In this paper a single rotated face location algorithm based on affine template matching is presented. An average frontal face is used as the base, from which all templates of various poses are produced by affine transforms varied in stretch and pose. The templates are divided into a series of pairs: eyes in whole and face itself. The eyes in whole templates are used first to search for possible eye regions and then the face templates are matched to get possible face region, finally some mosaic rules are used to verify face regions, the region corresponding to the maximum matching is output as the result. Experiments over two test sets with 144 images and 60 images have achieved a detection rate of 79.2% and 85% respectively. Those results demonstrate the feasibility of this approach.

关 键 词：人脸定位 人脸检测 仿射模板匹配 人脸图像处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]