针对东方人黑色虹膜识别的光谱选择  

Wavelength band selection for black-based iris recognition of East Asians

作　　者：宫雅卓[1,3] 张大鹏[2] 施鹏飞[1] 严京旗[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200240 [2]香港理工大学计算机系,生物识别研究中心,中国香港 [3]上海理工大学数字印刷研究所,上海200090

出　　处：《中国科学：信息科学》2012年第7期859-868,共10页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:61020106004);国家重点基础研究发展计划(批准号:2011CB302203);国家自然科学基金委员会及香港研究资助局联合研究计划;香港理工大学中央基金资助项目

摘　　要：本文以东方人黑色虹膜作为研究对象,进行两方面研究:1)对于单一光谱照明,找到某一个最适合东方人黑色虹膜的光谱段;2)对于多个光谱照明,确定东方人黑色虹膜所需要的光谱段数量.第一部分研究中,采用改进的卷积矩阵与频谱能量相结合的算法,对虹膜纹理质量进行评估;采用改进2-DGabor与1-DLog-Gabor相结合的编码方法,获得虹膜匹配性能的指标.实验显示:针对单一光谱照明,700nm对于东方人的黑色虹膜是最优的光谱段.第二部分研究中,采用改进multi-group(2D)2PCA算法,基于多光谱虹膜图像的最大不相关性进行层次聚类分析.实验显示:针对多个光谱照明,3个主聚类可以最好地描述从545nm到940nm的所有光谱段.以上研究在该领域内具有开拓性,解决了东方人黑色虹膜多光谱采集与识别的基础问题,为东方人黑色虹膜的多光谱融合提供了理论依据.In this work, the black iris of East Asians is captured at 12 wavelengths, from 420 to 940 nm. The purpose is （1） to find a band of the electromagnetic spectrum, in which more texture information can be extracted from the iris of East Asians, and （2） to determine how many bands of spectral wavelengths will be enough for black-based iris multispectral fusion and find these bands. In section I, the improved convolution matrix and spectrum energy are combined to measure the quality of iris image, and the improved 2-D Gabor wavelet and 1-D Log-Gabor wavelet are combined to evaluate the iris recognition accuracy. The experiments suggest that 700 nm is the most suitable wavelength for black iris recognition. In section II, using the agglomerative clustering based on an improved multi-group two-dimensional principal component analysis （（2D）2PCA）, we determine that 3 clusters are enough to represent the 10 feature bands of spectral wavelengths from 545 to 940 nm. This research represents the first attempt in the literature to investigate the iris of East Asians in a multispectral analysis, answered the fundamental questions about multispectral iris recognition of East Asians, and provided a theoretical basis for iris multispectral fusion.

关 键 词：图像聚类 虹膜识别 多光谱成像 图像融合 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]