摘要
利用巴氏距离(Bhattacharyya Distance)和PCA(Principal Component Analysis)相结合进行人脸识别研究,提出了使用巴氏距离和PCA相合的算法对特征进行提取。当特征向量维数高时,首先对样本K-L(Karhunen-Loeve)变换进行降维,然后采用巴氏距离特征的迭代算法,得到最小错误率上界。基于ORL人脸数据库的实验表明该方法的识别性能优于LDA、HPCA、HLDA,采用文中的算法可以有效地提高识别率,减少巴氏距离特征计算时间,具有较强的实用性。
This paper studies the face recognition based on the methods of Bhattacharyya distance and principal component analysis,and proposes a smart feature selection method which combines principal component analysis and Bhattacharyya distance.When the feature vector dimension is high,the sample dimension is reduced by using K-L decomposition.It gets the smallest error rate upper bound by using iterative algorithm which has Bhattacharyya distance feature.The experiments in ORL face database shows that its performance is better than using the methods of LDA,HPCA and HLDA.This algorithm can raise the recognition rate effectively and reduce the time which is used to calculate the Bhattacharyya distance.It has strong practicability.
出处
《计算机工程与应用》
CSCD
2012年第3期202-204,215,共4页
Computer Engineering and Applications
基金
广东省重大科技专项(No.2009A080202006)
广东省自然科学基金资助项目(No.9151009001000021)
广东省教育部产学研合作专项资金资助项目(No.2009B090300341)
关键词
巴氏距离
主分量分析
人脸识别
PCA和K-L变换相结合(PCA+K-L)
Bhattacharyya distance
Principal Component Analysis（PCA）
face recognition
PCA＋K-L（Principal Component Analysis and Karhunen-Loeve）
