手掌纹脉的自适应共轭鉴别能量分析认证算法  

Adaptive Conjugate Discrimination Power Analysis for Palm-print-vein Verification

作　　者：刘刚[1] 黎明[1,2] 冷璐[1,3] 汪宇玲[1,2] 

机构地区：[1]南昌航空大学无损检测技术教育部重点实验室,江西南昌330063 [2]南京航空航天大学自动化学院,江苏南京210016 [3]延世大学电气电子工程学院,首尔120749

出　　处：《计算机仿真》2015年第5期364-367,共4页Computer Simulation

基　　金：国家自然科学基金(61305010,61262019,61202112,61303199);韩国国家研究基金基础科学研究计划项目(2013006574);中国博士后科学基金(2013M531554);南昌航空大学博士启动基金(EA201308058)

摘　　要：在手掌纹脉图像识别的研究中,多模态融合可以多方面提高生物特征系统的性能。由于错位干扰,掌纹和掌脉的纹理特征编码难以实现双模态同时配准的特征级融合,也难以直接对特征数据进行筛选和压缩。采用离散余弦变换分别提取非接触式掌纹和掌脉特征,避免了匹配配准的平移操作。对特征空间共轭拓展,解决了特征维数和表征形式的兼容性问题。通过特征级融合有效保留了原始鉴别信息,并对特征进行有效归一化处理,实现了分量选择的自适应性。通过提出的自适应共轭鉴别能量分析算法对共轭特征分量进行排序,避免了掩膜窗口的优化问题,筛选出更高鉴别性的分量组合。通过与现有掩膜以及单模态认证方案的对比,验证了手掌纹脉图像特征融合算法在普适性、稳定性和认证精度等方面的有效性。Multi-modal fusion can enhance the performance of biometric systems in many ways. Due to the dislo- cation problem, texture feature codes of palmprint and palmvein can be neither used for the dual-modal alignment fu- sion at feature level, nor directly selected or compressed. The features of contactless palmprint and palmvein, which are extracted with discrete cosine transform （ DCT）, do not need to be muhi-translated for matching alignment. The feature space is conjugately expanded to solve the compatibility problem of the dimension and representation of fea- tures. The features are fused to effectively preserve raw discriminant information, and normalized to enhance the a- daptability of coefficient selection. The proposed adaptive conjugate discrimination power analysis （ACDPA） conju- gately orders the feature components, which can avoid the optimization of the pre-masking window, so that the combination of selected feature contains high discrimination. The comparison with the existing pre-masking-based and single-modal algorithms confirms the superiorities of ACDPA in terms of universality, stability and verification accuracy. KEYWORDS：Adaptive conjugate discrimination power analysis; Conjugate expansion of feature space; Contactless palm-print-vein verification

关 键 词：自适应共轭鉴别能量分析 特征空间共轭拓展 非接触手掌纹脉认证 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]