多特征融合的人脸表情识别  

Recognition of Facial Expression with Multi-Feature Fusion

作　　者：李春英[1] 汤志康[2] 黄春艳[1] 颜春萍[1] 张仙玉[1] 陈岚[1] 

机构地区：[1]肇庆学院计算机学院,广东肇庆526061 [2]广东技术师范学院计算机学院,广东广州510665

出　　处：《肇庆学院学报》2012年第2期19-24,共6页Journal of Zhaoqing University

摘　　要：提出了一种多特征融合的表情识别模型:首先,对预处理后的图像提取2种局部描述算子Gabor特征和多元中心化二值模式特征,根据对表情的贡献程度划分表情子区域;其次,通过主成份分析法对表情子区域的特征向量进行降维,并构建随机子空间训练分类器;最后,利用Bagging技术提高多分类器的分类性能,并采用加权投票的融合规则进行决策判别.人脸表情库的实验结果表明,此方法有很好的鲁棒性和识别率.This paper proposes a recognition model of facial expression with multi-feature fusion. In the preprocessing stage of images, two local description operators （Gabor feature and Multi-Scale Centralized Bi- nary Pattern （MCBP） feature） were extracted, and on the basis of their contribution to the recognition, sub-re- gions of expression were established. Then the vector dimensions of these sub-regions were reduced by PCA and the random subspace training classifiers were constructed. Finally, the performance of multiple classifiers was improved by the application of bagging technique, and the final decision was determined by a rule of merging weighting and voting. The experimental findings of the facial expression databases （JAFFE） validate that the proposed method not only has good robust but can also increase the accuracy of the system.

关 键 词：表情识别 局部特征 多特征融合 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]