基于SVM RFE的人脸特征选择方法  

Facial feature selection method based on SVM RFE

作　　者：李伟红[1] 龚卫国[1] 陈伟民[1] 梁毅雄[1] 尹克重[1] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室,重庆400044

出　　处：《光电工程》2006年第5期113-117,共5页Opto-Electronic Engineering

基　　金：国家教育部科学研究重点项目(02057);重庆市自然科学基金重点研究项目(CSTC2005BA2002);重庆市自然科学基金项目(CSTC2005BB2181)

摘　　要：提出一种新的基于SVMRFE(SupportVectorMachineRecursiveFeatureElimination)的人脸特征选择方法。该方法将权重矢量和半径/间隔作为SVMRFE的特征选择标准,采用缩放因子梯度算法优化特征搜索。基于该方法构建了一种实用、有效的人脸特征提取、选择及识别框架,并在UMIST人脸数据库上进行了验证实验。对特征选择前后的分类能力及速度进行了分析比较,结果表明,该方法是一种实用、有效的人脸特征选择方法,可以在特征维数为80左右时,达到94.62%的分类识别率。A novel method based on Support Vector Machine Recursive Feature Elimination （SVM RFE） for facial feature selection is proposed. The proposed method made weight vector and radius/margin to act as feature selection criteria and adopted scaling factor and gradient descent algorithm for optimizing the feature search performance. Based on the method, a practical and efficient framework of the facial feature selection was constructed. To compare the classification ability and performance of the proposed method, experiments were performed on UMIST facial database. An accuracy of 94.62 % is achieved when facial feature are around 80, which shows the efficiency of proposed method .

关 键 词：特征选择 人脸识别 支持向量机 SVM RFE 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]