基于Contourlet子带能量特征多HMM融合的静脉识别.pdfVIP

第33 卷第8 期 电 子 与 信 息 学 报 Vol.33No.8 2011 年8 月 Journal of Electronics Information Technology Aug. 2011 基于Contourlet 子带能量特征多HMM 融合的静脉识别 贾 旭* 薛定宇 崔建江 刘 晶 ( 东北大学信息科学与工程学院 沈阳 110819) 摘 要：为了准确识别人的身份，该文提出了一种以轮廓波(Contourlet)变换后不同尺度下的子带能量为特征，建 立并融合多个隐马尔科夫模型(HMM)的手背静脉识别算法。该算法首先采用了光强可调的近红外阵列光源，通过 逐步增加光强来获得手背静脉图像序列；而后，将每一静脉图像进行 Contourlet 变换，并计算不同尺度下每一子 带的能量，以3 个尺度下子带能量作为特征观测值建立3 个HMM；最后，融合3 个HMM 计算得到的观测值发生 概率，将融合结果与阈值作比较，从而完成静脉识别过程。实验结果表明，提出的算法可以使真实匹配与虚假匹配 的区分度最大化，与基于特征点或静脉信息融合的识别算法相比，正确识别率得到了提高。 关键词：特征提取；静脉识别；轮廓波变换；隐马尔科夫模型(HMM) 中图分类号：TP391.4 文献标识码： A 文章编号：1009-5896(2011)08-1877-06 DOI: 10.3724/SP.J.1146.2010.01253 Vein Recognition Based on Fusing Multi HMMs with Contourlet Subband Energy Observations Jia Xu Xue Ding-yu Cui Jian-jiang Liu Jing (School of Information Science Engineering, Northeastern University, Shenyang 110819, China) Abstract: In order to recognize one’s identity accurately, a dorsal hand vein recognition algorithm based on establishing and fusing multi Hidden Markov Models (HMMs) is proposed in the paper, where multi-scale subband energies are used as the features of HMMs after the vein images are processed by Contourlet transform. In the proposed algorithm near infrared light source array whose light intensity can be adjustable is applied, and the dorsal hand vein image sequence is acquired through increasing the light intensity gradually. Then every vein image is processed by Contourlet transform, subband energies under three scales are computed and used as the features of three HMMs. Finally, the probabilities of three HMMs generating observable s