掌静脉识别算法研究

发布时间：2018-08-06 18:38

【摘要】：随着科学技术的发展,安全问题越来越受到人们的关注。如何准确可靠地识别出一个人的身份,已经成为一个亟待解决的问题。传统的身份识别机制存在容易盗取和复制的缺点,已经不能满足人们对于高安全性的要求。生物特征识别技术是解决该问题的有效途径,其中,掌静脉识别技术是近年来兴起的一种生物特征识别技术,具有高防伪性、识别精度高和容易被用户接受等优点。本论文主要针对掌静脉识别算法进行了比较深入的研究,研究比较了ROI(Region of Interest,感兴趣区域)图像的获取和ROI图像的图像增强和去噪方式,重点研究了NBP(Neighbor based Binary Pattern,近邻二值模式)特征的提取匹配方法,SIFT(Scale Invariant Feature Transform,尺度不变特征)特征的提取匹配方法并提出了一种融合纹理特征和局部不变特征的掌静脉识别方法。本论文的主要工作和研究成果如下:1.就掌静脉识别ROI图像的获取和预处理进行了初步研究和比对实验,选取基于掌心矩形的ROI图像提取方式和CLAHE(Contrast Limited Adaptive histogram equalization,限制对比度自适应直方图均衡)方法与中值滤波的方法对图像进行图像增强和去噪,以达到后续试验的最好效果。2.研究了基于NBP特征和基于SIFT特征的两种单特征的掌静脉识别方法。针对SIFT匹配过程中传统RANSAC(Random Sample Consensus,随机抽样一致算法)方法的弊端,提出了一种基于相似度距离剔除错误匹配点的方法,提高错误匹配点剔除的效率及准确率,使SIFT算法在掌静脉识别中的应用更为准确。然后,设计实验对两种掌静脉识别方式进行比较分析,得到两种算法性能特点,为后续融合两种算法提供依据和思路。3.针对前文研究的两种算法的优缺点,通过对两种算法融合的可行性进行分析:NBP特征作为一种全局特征,SIFT作为一种局部特征,两者对于不同手掌图像的区分度和相同手掌图像的匹配度具有较强的互补性,并且在实验时间NBP特征也可以弥补SIFT特征的不具有实时性的劣势,在鲁棒性上SIFT特征可以弥补NBP特征在较大位移上鲁棒性差的劣势,得到两种算法非常适合进行信息融合的结论,提出一种融合纹理特征和局部不变特征的掌静脉识别算法,提高了掌静脉算法的识别正确率,在PolyU掌纹库和实验室自采库上分别取得正确识别率99.114%和99.722%的效果。
[Abstract]:With the development of science and technology, people pay more and more attention to the security problem. How to identify a person accurately and reliably has become an urgent problem. The traditional identification mechanism is easy to steal and copy, which can not meet the requirements of high security. Biometric recognition technology is an effective way to solve this problem. Among them, metacarpal vein recognition technology is a kind of biometric recognition technology developed in recent years, which has the advantages of high anti-counterfeiting, high recognition accuracy and easy to be accepted by users. This paper mainly focuses on the palmar vein recognition algorithm, studies and compares the ROI (Region of Interest, image acquisition and ROI image enhancement and denoising methods. In this paper, the extraction and matching method of NBP (Neighbor based Binary Pattern, nearest neighbor binary pattern) feature is studied, and a method of palmar vein recognition based on texture feature and local invariant feature is proposed, which is based on sift (Scale Invariant Feature Transform, scale invariant feature. The main work and research results of this thesis are as follows: 1. The acquisition and preprocessing of ROI images of metacarpal vein recognition were preliminarily studied and compared. The method of ROI image extraction based on palm rectangle and CLAHE (Contrast Limited Adaptive histogram equalization, restricted contrast adaptive histogram equalization method and median filter method are selected to enhance and de-noise the image so as to achieve the best effect of subsequent experiments. 2. Two methods of palmar vein recognition based on NBP feature and SIFT feature are studied. In view of the disadvantages of the traditional RANSAC (Random Sample Consensus, random sampling algorithm in SIFT matching process, a method based on similarity distance is proposed to eliminate the error matching points, which can improve the efficiency and accuracy of error matching points elimination. The application of SIFT algorithm in metacarpal vein recognition is more accurate. Then, the experiment is designed to compare and analyze the two methods of palmar vein recognition, and the performance characteristics of the two algorithms are obtained, which provide the basis and train of thought for the subsequent fusion of the two algorithms. In view of the advantages and disadvantages of the two algorithms mentioned above, this paper analyzes the feasibility of the fusion of the two algorithms by analyzing the SIFT as a global feature, and sift as a local feature by analyzing the feasibility of the fusion of the two algorithms. The two have strong complementarities for different palm image differentiation and the same palm image matching degree, and the NBP feature can also make up for the disadvantage of non-real-time SIFT feature in the experimental time. In terms of robustness, SIFT features can make up for the disadvantage of poor robustness of NBP features on large displacement. It is concluded that the two algorithms are very suitable for information fusion. A palmar vein recognition algorithm based on texture feature and local invariant feature is proposed. The recognition rate of metacarpal vein algorithm was improved, and the correct recognition rates were 99.114% and 99.722% in PolyU palmprint database and laboratory self-mining database, respectively.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】