面向安全认证的掌纹掌脉特征融合识别方法研究

The legality of customer identity in bank security authentication process is verified by remote operation, which put forward high requirement on real-time, security and accuracy. Compared with other biological characteristics, palmprint is difficult to be duplicated or counterfeited. Meanwhile, palm vein characteristics is safe because of non-duplication. However, palmprint and palm vein are both interlaced and uneven thickness, which lead to deformation easily. Thus, it is very difficult to recognize palmprint and palm vein for bank security certification. Firslty, some geometric and texture features with invariant attribute are extracted to meet those problems. The correlation between different features are analyzed. Furthermore, the independent palmprint and palm vein features are fused dynamically, and then, an original variable feature space is built. Secondly, a novel optimization objective function for feature space is constructed, and an optimization algorithms is studied to obtain the optimal feature space. After that, the original feature space is mapped to a joint three-dimensional space, and the continuous data are transformed into discrete data type. Thirdly, a novel clustering and search method for three-dimensional characteristics space is studied to obtain a candidate feature space subset. Besides that, the integrated weighted matching strategy for random region is used to achieve rapid identification. The proposed method will not only form a systemic recognition theory and method with palmprint and palm vein feature for bank safety authentication, but also provide a theoretical basis for other biometric characteristic recognition in bank security authentication. Furthermore, the research results will take bank safety authentication to a new level.

银行安全认证以远程方式验证客户身份的合法性，对数据传输的实时性、安全性及准确性具有很高要求。与其他生物特征相比，掌纹难以复制和仿造，掌脉具有不可复制、安全性高等特点，但掌纹掌脉相互交错，粗细不均，且易形变，使得面向银行安全认证的掌纹掌脉识别十分困难。本项目研究具有不变属性的掌纹掌脉图像几何与纹理特征提取方法，分析各个特征的相关性，动态融合各个独立的掌纹掌脉特征，形成可变异的掌纹掌脉特征空间；同时构造特征空间优化目标函数，研究掌纹掌脉特征空间搜索方法，形成最优特征空间；然后，将该特征空间映射为三维联合特征空间，并进行离散化处理；在此基础上，研究三维特征空间的聚类与搜索方法，获得候选特征空间子集，运用随机分块策略进行综合加权匹配，实现掌纹掌脉的快速识别。形成面向银行安全认证的掌纹掌脉特征识别理论与方法，为银行安全认证的生物特征识别提供理论依据，提高我国银行业的安全认证水平。

商务安全认证以远程方式验证客户身份的合法性，对数据传输的实时性、安全性及准确性具有很高要求。与其他生物特征相比，掌纹难以复制和仿造，掌脉具有不可复制、安全性高等特点，但掌纹掌脉相互交错，粗细不均，且易形变，使得面向商务安全认证的掌纹掌脉识别十分困难。本项目针对掌纹掌脉特征属性知识难以存储的问题，利用结构矩阵方法，构造掌纹掌脉图像的最大树，将掌纹掌脉属性知识存储在树节点中，利用枝剪策略对掌纹掌脉图像进行滤波，获得了清晰的掌纹掌脉线条形态。在获取大量掌纹图像基础上，引入仿生视觉方法，研究了掌纹掌脉的形态、空间位置关系表示方法，提出了掌纹掌脉图像的几何特征与纹理特征提取方法，能提取出比例、位置、旋转等不变属性的掌纹掌脉特征；扩展了单个仿生视觉变换方法，提出了双生物启发变换的掌纹掌脉特征提取方法，建立了正交互补的掌纹掌脉二维特征空间。在此基础上，研究了掌纹掌脉特征空间的加权融合方法，提出了基于距离匹配的掌纹掌脉搜索匹配方法；改进了传统掌纹识别流程，增加了多层分类和分类定位搜索策略，其基本思想是，先根据掌纹特征，利用多层LDA主题模型对掌纹进行分类，再根据分类结果将掌纹特征分组存入数据库，搜索待检测掌纹的特征，利用前述构造的多层LDA对其进行分类定位。将提出的理论方法应用于无人零售店身份认证过程，解决了商品交易过程中的身份识别问题，形成了一套在商务安全认证过程中能有效识别掌纹掌脉图像特征的理论与方法。