一种加权矩形提取表情特征方法.docVIP

一种加权矩形提取表情特征的方法 　　摘 要:表情特征提取是面部表情识别的一个关键步骤。针对目前特征提取效率低的情况,通过分析Gabor特征提取的性质和积分图像计算效率高的特点,提出一种利用积分图像技术和矩形模板计算面部特征点特征的方法,模板模拟Gabor的多尺度性,每个模板定义相应的权值,表情图像按照Gabor的各个方向旋转,使用旋转图像积分图和加权模板而不是在积分图上旋转模板提取特征点的特征值,最后将此特征值用于表情分类。实验结果表明,该方法在识别结果相当的情况下极大地提高了特征提取的效率。 　　关键词:加权矩形; 积分图像; 矩形模板; 特征提取; 表情识别 　　中图分类号:TP391.41文献标志码:A 　　文章编号:1001-3695(2010)06-2364-03 　　doi:10.3969/j.issn.1001??3695.2010.06.105 　　 　　Weighted rectangle method for expression feature extraction 　　 　　LIN Qing??1,2??, ZHAO Xiang1, HU Rui??rui1, ZHAN Yong??zhao1, WANG Shi??tong??2,3?? 　　 　　(1.College of Computer Science Communication Engineering, Jiangsu University, Zhenjiang Jiangsu 212013,China; 2. Dept. of Computer, Nanjing University of Science Technology, Nanjing 210094, China; 3. School of InformationTechnology, Jiangnan University, Wuxi Jiangsu 214122, China) 　　 　　Abstract:The expression feature extraction is a key step in facial expression recognition. In view of the existing low efficiency in feature extraction, by analyzing the Gabor qualities and the high computational efficiency of integral image, this paper pre?勃?sented?? a method that combined integral image with rectangle template to obtain the value of facial feature points. These templates with homologous weight simulated the multi??scale of Gabor and expression image rotated via every direction of Gabor. Obtained the values of feature points not by rotating templates on rotation integral image but the joint of weighted templates and rotation integral image. Finally used these feature values for expression classification. The experimental results show that this method improves efficiency while extracting facial feature and does not affect the result heavily. 　　Key words:weighted rectangle; integral image; rectangle templates; feature extraction; expression recognition 　　 　　0 引言?? 　　随着人机交互领域的发展,面部表情识别越来越成为模式识别领域的研究热点。人脸表情含有丰富的信息,它是除了语言之外重要的信息交流方式,实际上,表情也是最直接、最显著的信息交流方式。?? 　　表情特征提取是表情识别的一个关键步骤,特征提取的好坏对识别结果产生巨大的影响。目前,特征抽取方法主要分为基于表观的