摘要
在局部鉴别典型相关分析(LDCCA)的基础上,提出一种广义局部判别型典型相关分析算法(GLDCCA)。该算法在准则函数的内协方差矩阵中引入样本类别信息,使其提取的特征更有利于模式分类,采用核主成份分析解决小样本问题,克服传统PCA所受到的线性约束。在人工数据集以及ORL和Yale 2个人脸库上进行实验,结果表明,与CCA算法和LDCCA算法相比,GLDCCA算法具有更高的识别性能。
On the basis of the Locality Discriminative Canonical Correlation Analysis（LDCCA）, this paper proposes a new supervised learning algorithm called Generalized Locality Discriminative Canonical Correlation Analysis（GLDCCA） algorithm, which can utilize much effectively the class information of samples in the covariance matrix. Meanwhile, Kernel Principal Component Analysis（KPCA） is used to solve the small sample problem and avoid the linear constraint which PCA is subjected to. Experimental results on artificial data sets, facial database including ORL and Yale show that the proposed GLDCCA algorithm is superior to CCA, LDCCA in recognition performance.
出处
《计算机工程》
CAS
CSCD
2012年第7期161-163,167,共4页
Computer Engineering
基金
安徽省高校优秀青年人才基金资助项目(2009SQRZ171
2010SQRL192)
安徽省教育厅自然科学基金资助项目(KJ2009B121)
安徽省高校省级自然科学基金资助项目(KJ2012Z395)
关键词
人脸识别
鉴别信息
典型相关分析
特征维数
特征融合
内协方差矩阵
face recognition
discriminative information
Canonical Correlation Analysis（CCA）
feature dimension
feature fusion
within covariance matrix
