双树复小波变换结合LM神经网络的掌纹识别方案  

Palmprint Recognition Scheme Based on DT-CWT and LM Neural Network

作　　者：卜宇[1] 刘俊霞[2] 唐学军[1] 

机构地区：[1]新疆工程学院计算机工程系,乌鲁木齐830052 [2]新疆工程学院电气与信息工程系,乌鲁木齐830023

出　　处：《计算机系统应用》2016年第9期165-170,共6页Computer Systems & Applications

基　　金：新疆维吾尔自治区高校科研计划青年教师科研启动基金(XJEDU2014S074)

摘　　要：针对现有掌纹识别方案不能够很好的提取多分辨率特征的问题,提出一种基于双树复小波变换(DT-CWT)和Levenberg-Marquardt(LM)神经网络的掌纹识别方案.首先,将彩色手掌图像转换成灰度图像.然后,提取出手掌图像中的感兴趣区域(ROI),并构建成直方图.接着,利用DT-CWT进行6层小波分解并获得特征系数,分别计算特征系数的最大值、平均值和中值构建36维特征向量.最后,利用LM神经网络根据特征向量实现掌纹的识别分类.在CASIA数据库上的实验结果表明,相比其他几种较新的识别方案,提出的方案的具有更高的识别率和更少的识别时间.For the issues that the existing solutions for palmprint identification can＇t be very good to extract multi-resolution characteristic, this paper proposed a palmprint recognition scheme based on dual-tree complex wavelet transform（DT-CWT） and Levenberg-Marquardt（LM） neural network. Firstly, it convertsthe color image into gray image. Then, the region of interest（ROI） is extracted from the palm image, and constructed a histogram. Then, this scheme uses DT-CWT for 6 layers of wavelet decomposition and obtains the characteristic coefficients, and calculatesthe maximum value, average value and median value of the characteristic coefficients respectively. Finally, it uses LM neural network to make recognition and classification of palmprint. Experimental results on CASIA database show that the recognition rate of the proposed scheme has high recognition rate and lower recognition time.

关 键 词：掌纹识别 双树复小波变换 特征提取 LM神经网络 感兴趣区域(ROI) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]