利用引入瞳孔位置的Gabor小波网络进行人脸识别  

Face Recognition by Pupils' Locations-Introduced Gabor Wavelet Networks

作　　者：谢竞[1] 阮秋琦[1] 

机构地区：[1]北京交通大学信息科学研究所,北京100044

出　　处：《信号处理》2009年第2期223-227,共5页Journal of Signal Processing

基　　金：国家自然科学基金(No.60472033);国家973重点基础研究发展规划(No.2004CB318005);教育部博士点基金(No.20030004023)

摘　　要：Gabor小波网络能很好地提取图像特征和进行图像表达,本文提出将瞳孔位置信息引入到Gabor小波网络的人脸特征提取中以提高提取效率。该瞳孔位置信息用于两个方面,一是在网络优化时利用瞳孔位置构造T形的小波初始位置分布,使得在小波数目一定的情况下识别信息的提取更高效;二是在小波网络的参数再确定时由瞳孔位置提供定位信息从而大大简化求参步骤。本文采用Gabor小波网络提取出人脸特征后再用核联想记忆法进行分类。实验结果表明,瞳孔位置的利用提高了人脸特征提取的效率;此外,与欧氏距离、归一化互相关和最近特征线(NFL)这些方法相比,核联想记忆法具有更好的识别率。Gabor wavelet networks are efficient for feature extraction and image representation. This paper approaches introducing pupils＇ locations information into the process of face feature extraction implemented by Gabor wavelet networks in order to improve the efficiency of feature extraction. The pupils ＇ locations information was used in two ways. Firstly, it was used to form the T-shape initial lo- cation distribution of Gabor wavelets in network optimization phase for the purpose of extracting more useful feature for recognition given a certain amount of wavelets. Secondly,it was used as locating information in the reparameterization phase to greatly simplify the repa- rameterization procedure. After feature extraction by Gabor wavelet networks, this paper adopts the method of kernel associative memory （KAM） to classify the feature. The experimental result showed that utilizing pupils＇ locations information could obviously improve the efficiency of feature extraction, and that compared to methods of Euclidean Distance, normalized cross correlation and nearest feature line （NFL） ,KAM achieved a better recognition rate.

关 键 词：人脸识别 Gabor小波网络 瞳孔位置 核联想记忆 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]