【作者】 王云新；

【导师】 刘铁根；

【作者基本信息】 天津大学 ， 光学工程， 2009， 博士

【摘要】 随着现代网络和信息技术的快速发展,人们对社会信息的安全性要求日益提高,基于生物特征的智能身份鉴别方法逐渐受到广泛重视。在众多生物特征识别方法中,手背静脉识别和虹膜识别具有稳定性、防伪性和非侵犯性等优势,是生物特征识别的两大研究热点,可广泛应用于政府、军队、银行、电子商务等领域。其中,手背静脉识别是一种新兴的身份鉴别方法,其研究和应用尚处于起步阶段。虹膜识别技术在近年来得到了快速发展,相对比较成熟,但虹膜识别的核心算法仍需要进一步的研究。本文首先建立了基于手背静脉的身份识别原型系统,围绕这一目标设计实现了近红外静脉成像装置,建立了天津大学手背静脉图像数据库,并从多分辨率纹理特征、代数特征、局部SIFT(Scale Invariant Feature Transform)特征三个方面研究了手背静脉识别算法,对算法进行性能测试,并与现有的典型算法进行对比,验证了算法的可行性和有效性。其次对虹膜识别中的关键技术进行了研究,主要包括虹膜定位、眼皮定位以及特征提取等。最后初步探讨了基于手背静脉和虹膜两种生物特征在匹配层的数据融合。本文的创新性工作概括如下:1、提出了基于多分辨率纹理特征分析的手背静脉识别方法。利用小波变换描述静脉纹理信息变化,实验分析了小波函数对识别性能的影响,评估了算法对手背垂直位移和旋转的容忍度。2、提出了基于局部SIFT特征分析的手背静脉识别方法。利用SIFT特征描述手背静脉的梯度信息,并改进SIFT特征的匹配算法,从而进一步提高了识别性能。该算法对手背的平移和旋转不敏感,具有重要的实用意义。3、提出了基于图像抽样的快速虹膜定位方法。基本思想是在抽样图像中进行粗定位,在原分辨率图像中实现精定位。利用抽样图像去除大量干扰信息,降低计算复杂度,改善了算法实时性。4、提出了基于最大连通路径的眼皮定位方法。在确定眼皮检测区域并增强水平方向边缘后,通过标注边缘图像中的连通路径,搜索具有最大水平距离的连通路径作为眼皮边缘点实现眼皮分割,使定位速度有了很大提高。5、提出将手背静脉和虹膜两种生物特征相结合进行身份识别。分别利用D-S证据理论和支持向量机方法在匹配层进行了融合实验和分析,融合后识别性能得到了较大提高。更多还原

【Abstract】 With the rapid development of modern network and information technology, the safety requirements for social information are constantly increased, and personal identification based on biometrics has received extensive attention. In the biometrics family, hand vein and iris recognitions are characteristic by stability, security, and un-intrusion, which are two research focuses and can be widly applied to many fields such as government, military, bank and electronic commerce. Hand vein recognition is a newly emerging identification technology, and its study and application are at the preliminary stage. Iris recognition technology has developed fast and become mature in recent years, whereas further study about kernel algorithms is still needed.Firstly a hand vein recognition prototype system is developed. Centering on this goal, we design a near-infrared (NIR) hand vein imaging system and construct Tianjin Universty (TJU) hand vein database. Three hand vein recognition methods are developed based on the multi-resolution texture feature, algebraic feature and local SIFT feaure. The proposed recognition methods for hand vein are tested and compared with the classical methods, which demonstrate their feasibility and effectiveness. Then the key iris algorithms are studied including iris localization, eyelid localization and iris feature extraction. Eventually, the paper preliminarily discusses the multi-biometric fusion based on the hand vein and iris at the match score level.The major innovations of the dissertation are as follows:1. A hand vein recognition method based on multi-resolution texture feature analysis is proposed. Wavelet transform is applied to describe the vein texture variety, the influence of wavelet function on recognition performance is analyzed by experiments, and the tolerance to hand shift and rotation is evaluated.2. A hand vein recognition method based on local SIFT feature is developed. The SIFT feature is adopted to describe the gradient information of hand vein, and the improved matching method for SIFT features enhances the identification performance. This algorithm is not sensitive to hand shift and rotation, which makes it has important practical significance.3. A rapid iris localization algorithm based on image sampling is advanced. Its basic idea is to locate the iris coarsely in the sampled image, and achieve the exact localization in the image with original resolution. Much disturbing information can be removed to reduce the computation complexity, and the real-time performance is improved greatly.4. An eyelid detection method based on maximal connection path is developed. After determining the eyelid detection region and enhancing the horizontal boundary, the maximal connection path is searched as the eyelid edge points through labeling the connected paths in the edge image, and then the eyelid segmentation can be achieved. This method increases the detection speed largely.5. The hand vein and iris recognitions are combined for personal identification. The D-S evidence theory and support vector machines (SVM) are applied to fusion experiment and analysis at the match score level, and the recognition performance has been improved greatly.更多还原