【作者】 李强；

【导师】 裘正定；

【作者基本信息】 北京交通大学 ， 信号与信息处理， 2006， 博士

【摘要】 生物特征识别是利用人独特的生理与行为特征来进行自动身份鉴别的技术，是解决社会信息化、数字化、网络化发展中安全问题的首选方案。多模态生物特征识别利用多种生物特征，可以提高系统的准确率、抗噪性、普适性、抗假冒攻击性能，减小大库性能衰减程度，实现更为鲁棒的系统，近年来受到广泛的重视，新应用与新算法不断涌现。如何构建低成本、低数据库复杂度、高用户接受度的系统，深入研究多模态融合层次、策略、算法等已是新的热点问题。 本文提出的手部特征包括掌纹、指横纹与手形特征，在单一模态生物特征研究的基础上，总结已有的特征级融合算法并借鉴决策级融合思想，提出基于关系度量的特征级融合方法(RMF)。以实现实用系统为目标，对融合基础理论展开研究，优化手部特征融合系统并进行实验评估，主要内容如下： 1．提出手部特征的概念。使用手部特征进行身份鉴别符合人的传统观念，可在避免用户接受度、成本与数据库管理等各方面问题前提下，实现多模融合系统。为充分利用手部图像中的生物特征，本文通过掌纹、指横纹与手形融合实现系统，并将之定义为狭义的手部特征。在比较分析当前的多模数据库及以往掌纹、手形的采集方法的基础上，本文采用数码相机进行手部图像采集并建立HA-BJTU数据库。 2．在掌纹识别的研究中，论文从克服高维、小样本问题的角度出发，提出采用子空间分析方法提取掌纹特征：针对二维主成分分析特征仍处于高维空间的问题，提出进一步去除图像列相关，得到改进二维主成分分析(I2D-PCA)算法；针对小样本问题，提出使用正切子空间方法建立基于GMM的类内变化统计模型，并通过实验证实其有效性。在此基础上，使用I2D-PCA得到正切子空间，进一步提升识别准确率。更多还原

【Abstract】 Biometrics is the most secure and convenient way to satisfy the requirements for identity digitalization and virtualization in the coming network society, which refers to the automatic identification of an individual by using certain physiological or behavioral traits associated with the person. Multimodal biometric system is more robust than unimodal one. It takes advantages of multiple biometric traits to improve the performance in many aspects including accuracy, noise resistance, universality, spoof attacks, and reduce performance degradation in huge database applications. Recently, new algorithms and applications of multimodal biometrics are emerging rapidly. The research of building low cost, low database complexity, low intrusiveness system while developing different fusion strategy, algorithm in different fusion level is a promising new field and becoming the focus of recent studies.The goal of the thesis is to implement such a system by handmetric recognition based on feature level fusion strategy. The proposed "handmetric" refers to the fusion of palmprint, knuckleprint and hand geometry. First, we present several algorithms for palmprint and knuckleprint recognition respectively. Then, based on the investigation of the existing feature level fusion algorithms and the idea of well studied decision score level fusion method, a strategy for feature level fusion named "Relation Measurement Fusion (RMF)" is proposed. The final handmetric recognition system is implemented by RMF, and the evaluation results validate the effectiveness of it. Main content of this work is as follows:1. The definition of handmetric is presented. The use of handmetric accords with the traditional conception of "identity". While in the field of multimodal biometrics, handmetric system is designed to solve the problem of inconvenience, cost and database management by acquiring different evidences in the same hand image. An image database called "HA-BJTU" for further study is constructed using a CCD camera. From the images, we could obtain palmprint, knuckleprint and hand geometry更多还原