虹膜识别前处理算法的研究

发布时间：2018-09-19 05:53

【摘要】：在日益注重信息安全的今天,人们对身份认证有了新的理解,期望身份识别具有更高的安全性与准确性。传统意义上的身份识别手段不能有效地自动进行身份鉴别,并且存在诸多问题。在此背景下产生了基于生物特征的身份识别技术,其中虹膜识别技术凭借其各种特性成为该领域的研究重点。虹膜识别的前处理过程包括图像采集、图像质量评价、虹膜定位、归一化与图像增强等步骤。前处理的主要目的是为后续的特征提取与匹配提供高质量的输入,所以前处理的结果直接影响到整个虹膜识别系统的性能。本文主要研究虹膜识别的前处理算法,通过深入分析其中的图像质量评价方法与虹膜定位方法,来提升虹膜识别前处理方法的精确度与效率。具体工作内容如下:1.提出了一种基于BP神经网络的虹膜图像质量评价算法。本文通过三个处理过程对虹膜图像的质量进行系统的评价,首先对被采集图像进行活体检测,一旦发现为伪造图像则立刻剔除,该步骤可以大大提升系统的安全性,避免不必要的时间浪费。之后对图像质量进行粗评估,文中粗评价指标的计算复杂度较低,可以快速有效的初步筛选图像,为虹膜图像质量的精评估提供可靠输入。接着进行虹膜图像质量的精评估,本文分析了一些传统指标的不足,改进了这些指标的计算方法,设置了本文的精评价指标,最后通过BP(Back Propagation)神经网络对这些评价因子进行融合,得出分类集合。2.提出了一种改进的虹膜定位方法。该算法将虹膜定位过程分为虹膜内圆粗、精定位与虹膜外圆粗、精定位等步骤。对于虹膜内圆定位:首先将虹膜图像二值化与去噪,对瞳孔区域进行投影得到虹膜内圆的粗定位,该粗定位范围有效隔绝了眼睑和睫毛等噪声的干扰,又将瞳孔部分完全包围在其中,然后用Canny算子检测瞳孔边缘,再用最小二乘法进行内圆边界拟合,从而完成虹膜内圆的精定位。对于外圆定位:首先根据内圆定位结果,设立检测模板,从瞳孔的上下左右四个方向检测外圆边界点,根据这些边界点得到外圆的粗定位。然后根据该粗定位参数,缩小使用Daugman圆模板检测时圆心和半径的增量范围,从而大大减小计算量,得到外圆的精定位。3.对本文的前处理整体算法进行了实验验证。本文从虹膜识别系统的整体性能上来对提出的前处理算法进行实验验证,通过实验得出:本文的前处理算法使识别系统具有较高的正确识别率,提升了虹膜识别系统的性能。同时也对本文的质量评价方法与虹膜定位方法进行了单独的实验,说明了该两种方法和其同类算法相比具有的优势。综上所述,本文对虹膜识别的前处理算法进行了研究,提出了一种基于BP神经网络的质量评价方法,以及一种基于小范围搜索、最小二乘曲线拟合与Daugman圆模板检测结合的虹膜定位方法。
[Abstract]:Nowadays, people pay more and more attention to information security, people have a new understanding of identity authentication, and expect identity recognition to have higher security and accuracy. Traditional means of identity identification can not effectively identify automatically, and there are many problems. Under this background, biometrics based identity recognition technology has been developed, in which iris recognition technology has become the focus of this field by virtue of its various characteristics. The pre-processing process of iris recognition includes image acquisition, image quality evaluation, iris location, normalization and image enhancement. The main purpose of pre-processing is to provide high-quality input for subsequent feature extraction and matching, so the result of pre-processing has a direct impact on the performance of the whole iris recognition system. This paper mainly studies the pre-processing algorithm of iris recognition, through in-depth analysis of image quality evaluation and iris location methods, to improve the accuracy and efficiency of iris recognition pre-processing methods. The details of the work are as follows: 1. An iris image quality evaluation algorithm based on BP neural network is proposed. In this paper, the quality of the iris image is systematically evaluated through three processing processes. Firstly, the captured image is detected in vivo. Once the false image is found, it is eliminated immediately, which can greatly enhance the security of the system. Avoid unnecessary waste of time. After rough evaluation of the image quality, the calculation complexity of the rough evaluation index is low, which can quickly and effectively screen the image, and provide reliable input for the accurate evaluation of the iris image quality. Then the iris image quality is evaluated carefully. The shortcomings of some traditional indexes are analyzed, the calculation method of these indexes is improved, and the fine evaluation indexes are set up. Finally, the BP (Back Propagation) neural network is used to fuse these evaluation factors. The classification set. 2. An improved iris location method is proposed. The algorithm divides the process of iris localization into three steps: inner circle of iris, fine localization and outer circle of iris, fine location of iris. For the location of iris circle: firstly, the iris image is binarized and denoised, and the coarse location of the iris circle is obtained by projecting the pupil area, which effectively insulates the interference of eyelids and eyelashes. The pupillary part is completely surrounded, then the pupil edge is detected by Canny operator, and the inner circle boundary is fitted by the least square method, so that the precise location of the iris inner circle is completed. For outer circle location: firstly, according to the result of inner circle localization, the detection template is set up, and the outer circle boundary points are detected from the upper and lower directions of the pupil, and the rough location of the outer circle is obtained according to these boundary points. Then, according to the coarse positioning parameters, the increment range of the center and radius of the circle detected by using Daugman circle template is reduced, and the calculation amount is greatly reduced, and the precise location of the outer circle is obtained. 3. Experimental verification of the whole pre-processing algorithm is carried out. In this paper, the proposed pre-processing algorithm is experimentally verified from the overall performance of the iris recognition system. The experimental results show that the pre-processing algorithm in this paper makes the recognition system have a higher recognition rate and improves the performance of the iris recognition system. At the same time, a separate experiment on the quality evaluation method and iris localization method is carried out in this paper, which shows the advantages of the two methods compared with the similar algorithms. To sum up, the pre-processing algorithm of iris recognition is studied in this paper, and a quality evaluation method based on BP neural network and a method based on small range search are proposed. Iris location method combined with least square curve fitting and Daugman circular template detection.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】