基于ICA的多姿态人脸表示  

Pose-varied Face Representation Using the Independent Component Analysis

作　　者：王刚[1] 刘伟[1] 冯贵玉[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,湖南长沙410073

出　　处：《国防科技大学学报》2003年第3期84-87,共4页Journal of National University of Defense Technology

基　　金：国家863高技术基金资助项目(2001AA114180);国家杰出青年科学基金资助项目(60225015);高等学校优秀青年教师教学科研奖励计划资助项目

摘　　要：将独立成分分析(ICA)应用于多姿态人脸识别。对比分析了ICA和主成分分析(PCA)两种人脸识别方法的差异,并重点研究了多姿态人脸的独立成分(IC)表示。在基于权向量幅值的方法基础上,引入了基于比例因子的IC核选择的新方法。实验表明,新方法有利于提高识别的准确率和识别的效率。Independence component analysis (ICA) is applied in posevaried face recognition. Discriminations between ICA and principal component analysis (PCA) in face recognition are analyzed, and independent component (IC) representation in posevaried face is discussed in detail. Based on the method that selects a subset as the kernel for the representation by ordering the sources via the magnitude of the corresponding weights, a novel IC representation of posevaried face based on the scale factor is proposed. Demonstration indicates that the proposed method is efficient.

关 键 词：独立成分分析 多姿态 人脸表示 比例因子 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]