图像的充分矢量三角形模式.doc

中图法分类号：TP391 文献标识码：A 文章编号： 精确局部特征描述的表情识别 胡敏1,2，江河1,2，王晓华1,2，陈红波1,2，李堃1,2，任福继 1,2 （1合肥工业大学计算机与信息学院 合肥 230009 2合肥工业大学情感计算与先进智能机器安徽省重点实验室 合肥 230009） 摘 要：目的 针对传统局部特征提取方法在表情识别上的局限性，提出一种精确局部特征描述的表情识别方法。方法 首先将人的眉毛、眼睛和嘴巴3个对表情识别起关键作用的器官分割出来，使得特征描述更具有针对性。然后，构造充分矢量三角形以统计图像的轮廓特征与细节特征。最后，对于不同的表情器官采用不同尺度的充分矢量三角形描述，对于同种表情器官采用多种尺度的充分矢量三角形联合描述，从而充分描述关键器官的图像信息。结果 该算法在日本女性表情人脸库(JAFFE)、cohn-kanade库(CK)以及Pain Expressions库上进行实验，分别取得了95.67%、97.83%、84.0%的平均识别率以及11.70ms、30.23ms、11.73ms的平均特征提取时间，实验结果表明精确局部特征描述的表情识别方法可以较快、较准确的进行人脸表情识别。结论 精确局部特征描述的表情识别方法通过器官的分割以及充分矢量三角形模式的构造与灵活运用，良好的表达了图像的局部特征且具有较低的时间复杂度，将本文算法与目前典型的表情识别算法的实验对比结果也证明了本文算法的有效性。 关键词：表情识别；精确局部特征；充分矢量三角形模式；多种尺度 Facial expression recognition of precise local feature description Hu Min1,2, Jiang He1,2, Wang Xiahua1,2, Chen Hongbo1,2, Li Kun1,2, Ren Fuji1,2 (1.School of Computer and Information of Hefei University of Technology, 2.Anhui Province Key Laboratory of Affective Computing and Advanced Intelligent Machine, Hefei 230009, China) Abstract: Objective In order to identify facial emotions more accurately, this paper proposes a facial expression recognition method of precise local feature description. Methods Firstly, eyebrows, eyes and mouth are divided from a facial expression image and then the local features are gotten from these organ images which are processed by the expanded vector triangle pattern, thus the outline features of the image and the detail features of the image can be statistic respectively. Finally, different scales of sufficient vector triangle patterns are used to describe the features of different organs, and a variety of scales of sufficient vector triangle patterns are combined together to describe the same organ’s features, so that the information in the key organs can be expressed more fully. Results Experiments of the proposed method are performed on JAFEE, Cohn-Kanade(CK) and Pain Expressions database, it achieved an average recognition rate of 95.67%, 97.83%, 84.0% and an average time of featur