基于局部人脸图像的ICA人脸识别方法.pdfVIP

( ) 计算机工程与设计 Computer Engineering and Design 2550 2010,31 11 计算机工程与设计 Computer Engineering and Design 人工智能 基于局部人脸图像的ICA 人脸识别方法 黄 璞， 陈才扣 (扬州大学信息工程学院，江苏扬州225009) 摘 要：提出了一种基于局部人脸图像独立分量分析的特征提取方法。该方法将人脸图像分成若干个相等的部分，将分成 的局部人脸图像矩阵作为训练样本，并先后从水平方向、垂直方向提取训练样本的独立分量。相较于传统的独立分量分析 (ICA) 方法，该方法具有如下优点：有效解决了传统ICA 在进行特征抽取过程中的高维小样本问题；将局部人脸图像作为训 练样本，这不仅增加了训练样本数，而且有利于提取人脸局部特征；依次从训练样本的水平方向、垂直方向提取训练样本特 征，使得提取的特征不仅维数更小，而且能更有效地反映样本的局部信息。以上优点使得提出的算法较传统方法在人脸识 别方面更稳定，识别率更高，在Yale 人脸库和AR 人脸库上验证了该算法的有效性。 ( ) ( ) 关键词：分块; 独立分量分析 ICA ; 主分量分析 PCA ; 局部特征; 特征提取; 人脸识别 中图法分类号：TP391 文献标识码：A 文章编号：1000-7024 (2010) 11-2550-04 ICA face recognition method based on partial face image HUANG Pu, CHEN Cai-kou (College of Information Engineering, Yangzhou University, Yangzhou 225009, China) Abstract ：A feature extraction method based on partial face image independent component analysis is proposed. This method partitions the facial image into blocks firstly, then takes partial face image matrix as the training sample, and extracts horizontal and vertical in- ( ) dependent components from the training samples sequentially. Compared with traditional independent component analysis ICA , the developed method has the following advantages: it can effectively solve the problem of ICA in the process of feature extraction, i.e. the number of available training sample is great less than that of the training vector; it takes partial face image as the training sample which doesn ’t only increase the number of samples, but also is helpful t