增强的典型相关分析及其在人脸识别特征融合中的应用  

Enhanced CCA and its Applications in Feature Fusion of Face Recognition

作　　者：赵松[1] 张志坚[1] 张培仁[1] 

机构地区：[1]中国科学技术大学自动化系,合肥230027

出　　处：《计算机辅助设计与图形学学报》2009年第3期394-399,共6页Journal of Computer-Aided Design & Computer Graphics

摘　　要：在传统的典型相关分析(CCA)基础上,定义了类别相关性,提出了增强典型相关分析(ECCA)方法.对于一个模式空间的2个观测空间(对任意模式都有2种观测向量),ECCA能够找到这2个观测空间对类别而言更有意义的相关子空间,且同时保持了投影分量的无关性.实验结果表明,ECCA优于CCA,GCCA融合方法.Based on the theory of classical canonical correlation analysis （CCA）, by defining class correlation, an enhanced canonical correlation analysis （ECCA） is proposed. If a pattern space has two observation spaces （For any pattern, there are two observation vectors of different kinds）, ECCA can find the relevant subspaces of the observation spaces, in which the projections of original random vectors are irrelevant. Experiments demonstrate the superior performance of ECCA fusion over feature fusion algorithm such as CCA, GCCA.

关 键 词：增强典型相关分析 人脸识别 特征融合 异质图像融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]