基于融合特征提取与LLE方法的表情识别  

Expression Recognition Based on Fusion Features Extraction and LLE Method

作　　者：兰兰[1] 陈万忠[1] 魏庭松 LAN Lan CHEN Wanzhong WEI Tingsong(College of Communication Engineering, Jilin University, Changchun 130022, China)

机构地区：[1]吉林大学通信工程学院,长春130022

出　　处：《吉林大学学报（信息科学版）》2017年第4期384-391,共8页Journal of Jilin University（Information Science Edition）

基　　金：吉林省科技发展计划自然基金资助项目(20150101191JC)

摘　　要：为保证所提取特征表征作用的全面性,提出一种基于几何特征和局部纹理特征相结合的特征提取方法。将基于主动表观模型(AAM:Active Appearance Model)特征点标记提取的几何特征和基于局部二值模式(LBP:Local Binary Pattern)提取的眼部和嘴部纹理特征进行融合,融合后的特征经局部线性嵌入(LLE:Locally Linear Embedding)方法进行特征降维,并使用多分类的支持向量机(SVM:Support Vector Machine)进行分类识别。该方法分别选取JAFFE数据集7类表情和小样本数据集Yale的4类表情进行实验,识别准确率分别达到了98.57%和91.67%,从而证明了该方法的有效性。Feature extraction is a basis, a vital step and a major issue in facial expression recognition. To ensure that the extracted features can be more comprehensive characterization of a certain kind of expression, we present a feature extraction method based on fused geometry and local texture features. Geometric features are obtained from the feature points marked by AAM （Active Appearance Model） algorithm, texture feature extraction is based on LBP （ Local Binary Pattern） algorithm, the dimension of fusion expression features is reduced by LLE （Locally Linear Embedding ） algorithm. Finally, a multi-class SVM （Support Vector Machine） is used for facial expression classification. Our method is deployed on the JAFFE and Yale data sets, the results show a recognition accuracy of 98.57% and 91.67% respectively, which prove the effectiveness of our proposed method.

关 键 词：表情识别 主动表观模型 局部二值模式 局部线性嵌入 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]