基于纠错输出编码的人脸表情识别  

Facial expression recognition based on error-correcting output coding

作　　者：余棉水[1] 朱岸青[1] 解晓萌[2] 

机构地区：[1]广东工贸职业技术学院计算机系,广州510510 [2]华南理工大学计算机科学与工程学院,广州510641

出　　处：《计算机工程与应用》2014年第3期155-159,共5页Computer Engineering and Applications

基　　金：广东省产学研合作项目(No.2012B091100043)

摘　　要：多分类问题一直是模式识别领域的一个热点,提出了一种基于纠错输出编码和支持向量机的多分类器算法。根据通信编码理论设计纠错输出编码矩阵;按照该编码矩阵设计若干个互不相关的子支持向量机,根据编码原理将它们融合为一个多分类器。为了验证本分类器的有效性,采用Gabor小波提取人脸表情特征,应用二元主成分(2DPCA)分析法对提取的特征进行降维处理,应用该分类器进行了人脸表情的识别。实验结果表明,提出的方法能有效提高人脸表情的识别率,并具有极好的鲁棒性。Multiple classification problems has been a hot topic in the field of pattern recognition. This paper proposes a multiple classifier algorithm based on Error-Correcting Output Coding（ECOC）and Support Vector Machine（SVM）. According to the communication coding theory to design error-correcting output coding matrix, it constructs some irrelevant SVMs, and integrates them as a multiple classifier. In order to verify the effectiveness of the classifier, using the Gabor wavelet to extract facial expression features, and application of two principal components（2DPCA）to reduce the dimension of extracted features, the classifier is used for facial expression recognition. Experimental results show that the method can effectively improve facial expression recognition and has excellent robustness.

关 键 词：支持向量机 多分类器 纠错输出编码 GABOR小波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]