基于多层次分类策略的非特定人人脸表情分类识别  

Person-independent facial expression recognition based on multi-level classification strategy

作　　者：胡步发[1] 陈炳兴[1] 黄银成[1] 

机构地区：[1]福州大学机械工程及自动化学院,福州350108

出　　处：《计算机应用》2010年第12期90-94,共5页journal of Computer Applications

基　　金：福州大学科技创新基金资助项目(2008-XQ-15)

摘　　要：针对目前非特定人人脸表情平均识别率普遍不高这一问题,提出了一种新的人脸表情分类识别方法。该方法基于多层次分类策略,在第一层分类阶段,首先使用局部二值模式(LBP)提取人脸全局特征,然后使用主元分析(PCA)方法降低特征维数,使用支持向量机(SVM)方法对全部表情进行第一次分类;在第二、第三层分类阶段,将人脸表情特征差异区域进行分块,采用LBP对各个子块提取特征并分配相应权值,突出有用特征信息,然后使用PCA方法降低特征维数并用欧氏距离方法分类。在JAFFE人脸表情数据库中进行实验,获得了74.76%的平均识别率。实验结果表明,该方法与其他方法相比具有更好的整体泛化性能和更高的平均识别率。In this paper,a novel method of facial expression recognition was proposed to solve the problem that the average recognition rate of person-independent facial expression is not high at present.The method was based on the multi-level classification strategy.In the stage of the first classification,global features were firstly extracted by Local Binary Pattern（LBP）,then Principal Component Analysis（PCA） method was used to reduce the feature dimension,and Support Vector Machin（SVM） classifier was used to classify the expressions.Then in the stages of the second and the third classification,each local differential feature＇s image of facial expression was partitioned into several regions,each region feature was extracted by LBP and assigned a weight,the useful features were enhanced,then PCA method was used to reduce the feature dimension,and Euclidean distance method for classifying was adopted.The experiments were conducted on Japanese Female Facial Expression（JAFFE） database and the average recognition rate of 74.76% was achieved.Theoretical analysis and experimental results show that the proposed method has better generalization performance and higher average recognition rate than other methods.

关 键 词：人脸表情 非特定人 多层次分类 局部二值模式 支持向量机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]