一种基于局部特征融合的表情识别方法  

An expression recognition method based on local feature fusion

作　　者：邓捷方[1] 杨勇[1] 王国胤[1] 

机构地区：[1]重庆邮电大学计算机科学与技术研究所,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2011年第5期626-630,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金(60773113);重庆市杰出青年基金(2008BA2041);重庆邮电大学科研基金(A2009-26);重庆市计算机网络与通信技术重点实验室开放基金(CY-CNCL-2009-2)~~

摘　　要：表情识别是人工智能和模式识别的研究热点,而特征融合方法则是表情识别中重要的技术方法之一。基于嘴部的Gabor小波特征和几何特征对表情识别有重要作用,提出一种仅用嘴部不同特征进行特征融合的表情识别方法。该方法将嘴部的Gabor小波特征和几何特征进行特征融合后再使用最近邻分类器分类。根据不同样本库、不同识别方法的对比实验结果表明,该方法相对于单独的Gabor小波特征和单独的几何特征用于表情识别能得到更好的识别效果,且相对于整幅人脸表情图像的识别具有更好的实时性。Expression recognition is a hot topic in artificial intelligence and pattern recognition. However, feature fusion is one of the most important research subjects in expression recognition. In this paper, an expression recognition method based on serial feature fusion between Gabor wavelet feature and geometric feature of mouth is proposed. And the nearest neighbo~ rule is used to classify to get the comparison from different facial expression databases and different recognition methods. The experiment results show that this method obtains better recognition effect than any other methods in local feature expres- sion recognition. Moreover, it is proved that it can recognize express faster than global facial expression recognition.

关 键 词：表情识别 特征融合 GABOR小波 几何特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]