基于改进主动表观模型算法的人脸特征定位  

Facial feature points localization based on improved active appearance models algorithm

作　　者：黄飞[1] 谭守标[1] 

机构地区：[1]安徽大学电子信息工程学院,合肥230601

出　　处：《计算机工程与应用》2015年第16期204-209,共6页Computer Engineering and Applications

基　　金："核高基"国家科技重大专项分课题(No.20092X01031-001-004-04)

摘　　要：人脸特征点的精确定位一直是人脸图像处理的重要研究内容,特征点定位精确与否直接影响后续工作结果的好坏。在基于反向组合AAM(Active Appearance Models)人脸特征点定位算法的基础上,提出结合特征点局部纹理模型来对AAM初始形状参数做最优化以及对AAM匹配模板升级的改进。改进的算法采用特征点局部纹理模型和AAM全局纹理模型结合的方法来最优化AAM初始形状参数,并在此前提下对AAM匹配模板进行升级,使其更接近待匹配图像的信息。在精确的匹配模板和优化的初始形状参数下,匹配的最终精度会得到提升。实验和理论证明,改进后的算法比传统反向组合AAM算法以及现有改进的PAAM(Progressive AAM)算法以及简单的结合ASM和AAM的改进算法都有更好的特征点定位精度。The precise localization of face feature points is always an important research contents in face image processing,the accuracy of the face feature localization has a directly affect of the result of the follow-up work.In order to achieve a high-precision of the face feature localization,the local texture models is introduced to optimize the initial parameter of Active Appearance Models（AAM）and the upgrade of the matching template of AAM after the study of AAM-reversed algorithm.To optimize the initial parameter,the local appearance model is combined with the global texture model of AAM,the matching template of AAM is updated to make sure that it＇s closer to the real-time image.With the more precise template and the optimized parameter,the precise localization of face feature will be improved.It has been improved that algorithm has a better accuracy than the traditional AAM-reversed algorithm and the Progressive AAM（PAAM）algorithm through the theory and experimental.

关 键 词：人脸特征定位 主动表观模型 局部纹理模型 初始形状参数最优化 匹配模板升级 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]