基于遗传算法的虹膜识别技术研究与改进
发布时间：2018-11-24 12:58
【摘要】：虹膜识别技术因虹膜的优秀生物特性,在众多的身份鉴别技术中脱颖而出,被列为最为安全与精确的身份鉴别技术,具有广阔的应用前景与重要的学术研究价值。由于虹膜识别技术应用环境的复杂性以及其涉及领域的广泛性,其关键技术仍存在需要改进之处。本文结合虹膜图像自身的属性以及常见的虹膜识别系统流程,对虹膜定位、虹膜特征提取、虹膜特征降维等相关技术进行了系统的分析与研究。主要的工作如下：对Canny算子与Hou gh变换相结合的虹膜定位分割模型进行研究,针对传统Canny算子在边缘信息提取时存在容易受噪声影响以及需要手动输入阈值的缺陷,提出改进的Canny算子：首先利用S ober算子计算像素点的梯度幅值和方向,然后通过双线性插值求得梯度方向上的像素点幅值完成非极大值抑制,最后采用Otsu实现阈值自适应选取。利用改进的Canny算法与Hough变换结合实现对虹膜的定位,提升了定位的精确度。对定位后的虹膜图像利用坐标变换进行归一化处理并增强,完成虹膜图像预处理。针对基于2D-Gabor滤波器的虹膜特征提取得到的特征向量信息过于冗余的缺陷,提出了结合遗传算法的虹膜特征筛选模型,该模型实现了对虹膜特征向量的有效降维。对基于标准遗传算法实现的虹膜特征筛选模型进行研究,针对其中存在的缺陷,结合粒子群算法的优点,提出改进的遗传算法：在整体框架中融入粒子群算法,同时设计具有自适应性的遗传算子。利用改进的遗传算法对特征向量进行特征筛选,得到有效且低维的特征向量。最后采用移位Hamming距离差完成虹膜的分类,经过特征筛选的低维特征向量得到了更高的匹配准确率。本文实验的原始数据来自CASIA-V4-Thousand和CASIA-Iris-Lamp数据库,以衡量虹膜识别系统性能的评价标准False Accept Rate、False Reject Rate、Correct Recognition Rate、Equal Error Rate和Receiver Operating Characteristic Curve对系统进行测试,验证了本文提出的改进算法的有效性。
[Abstract]:Iris recognition technology has been listed as the safest and most accurate identification technology because of its excellent biological characteristics. It has broad application prospects and important academic research value. Due to the complexity of the application environment of iris recognition technology and its wide range of fields, the key technologies still need to be improved. In this paper, the iris location, iris feature extraction, iris feature dimensionality reduction and other related techniques are systematically analyzed and studied based on the iris image attributes and the common iris recognition system flow. The main work is as follows: the iris location segmentation model combined with Canny operator and Hou gh transform is studied. The traditional Canny operator is easy to be affected by noise and needs manual input threshold when extracting edge information. An improved Canny operator is proposed: firstly, S ober operator is used to calculate the gradient amplitude and direction of pixel points, then bilinear interpolation is used to obtain the non-maximum suppression of the pixel amplitude in the gradient direction. Finally, Otsu is used to adaptively select the threshold value. The improved Canny algorithm is combined with the Hough transform to realize the iris localization, which improves the accuracy of the location. The iris image is normalized and enhanced by coordinate transformation, and the iris image preprocessing is completed. Aiming at the defects of redundant information of iris feature extraction based on 2D-Gabor filter, an iris feature selection model combined with genetic algorithm is proposed, which can effectively reduce the dimension of iris feature vector. The iris feature screening model based on standard genetic algorithm is studied. Considering the shortcomings of particle swarm optimization algorithm and the advantages of particle swarm optimization algorithm, an improved genetic algorithm is proposed: integrating particle swarm optimization algorithm into the whole framework. At the same time, genetic operators with adaptability are designed. The improved genetic algorithm is used to screen the feature vectors and obtain the effective and low-dimensional feature vectors. Finally, the classification of iris is accomplished by shift Hamming distance difference, and the low dimensional feature vector which is filtered by feature can get higher matching accuracy. In this paper, the original data from CASIA-V4-Thousand and CASIA-Iris-Lamp database are used to measure the performance of iris recognition system. False Accept Rate,False Reject Rate,Correct Recognition Rate,Equal Error Rate and Receiver Operating Characteristic Curve are used to test the system. The effectiveness of the proposed improved algorithm is verified.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP18
本文编号：2353800
[Abstract]:Iris recognition technology has been listed as the safest and most accurate identification technology because of its excellent biological characteristics. It has broad application prospects and important academic research value. Due to the complexity of the application environment of iris recognition technology and its wide range of fields, the key technologies still need to be improved. In this paper, the iris location, iris feature extraction, iris feature dimensionality reduction and other related techniques are systematically analyzed and studied based on the iris image attributes and the common iris recognition system flow. The main work is as follows: the iris location segmentation model combined with Canny operator and Hou gh transform is studied. The traditional Canny operator is easy to be affected by noise and needs manual input threshold when extracting edge information. An improved Canny operator is proposed: firstly, S ober operator is used to calculate the gradient amplitude and direction of pixel points, then bilinear interpolation is used to obtain the non-maximum suppression of the pixel amplitude in the gradient direction. Finally, Otsu is used to adaptively select the threshold value. The improved Canny algorithm is combined with the Hough transform to realize the iris localization, which improves the accuracy of the location. The iris image is normalized and enhanced by coordinate transformation, and the iris image preprocessing is completed. Aiming at the defects of redundant information of iris feature extraction based on 2D-Gabor filter, an iris feature selection model combined with genetic algorithm is proposed, which can effectively reduce the dimension of iris feature vector. The iris feature screening model based on standard genetic algorithm is studied. Considering the shortcomings of particle swarm optimization algorithm and the advantages of particle swarm optimization algorithm, an improved genetic algorithm is proposed: integrating particle swarm optimization algorithm into the whole framework. At the same time, genetic operators with adaptability are designed. The improved genetic algorithm is used to screen the feature vectors and obtain the effective and low-dimensional feature vectors. Finally, the classification of iris is accomplished by shift Hamming distance difference, and the low dimensional feature vector which is filtered by feature can get higher matching accuracy. In this paper, the original data from CASIA-V4-Thousand and CASIA-Iris-Lamp database are used to measure the performance of iris recognition system. False Accept Rate,False Reject Rate,Correct Recognition Rate,Equal Error Rate and Receiver Operating Characteristic Curve are used to test the system. The effectiveness of the proposed improved algorithm is verified.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP18
【参考文献】
相关期刊论文 前10条
1 程宇奇;方伟;葛微;;抗噪的移位Hamming距离差虹膜匹配方法[J];模式识别与人工智能;2013年01期
2 毛巨勇;;2012年生物识别市场发展态势分析[J];中国安防;2012年12期
3 王博;;生物识别技术在司法领域的创新与发展[J];中国安防;2012年10期
4 武卫;;生物识别技术在机场安保领域的应用[J];中国民用航空;2012年08期
5 曹道友;程家兴;;基于改进的选择算子和交叉算子的遗传算法[J];计算机技术与发展;2010年02期
6 冯新岗;周诠;;数字图像中基于多尺度几何分析的圆检测算法[J];中国图象图形学报;2009年05期
7 苑玮琦;刘汪澜;柯丽;徐露;;用于虹膜特征提取的Gabor滤波器参数选取方法的研究[J];光电子.激光;2008年09期
8 王森;杨克俭;;基于双线性插值的图像缩放算法的研究与实现[J];自动化技术与应用;2008年07期
9 吴作好;曾洁;邹娟;杨晓东;张尧;;几种人体生物特征的生物识别技术比较[J];现代电子技术;2007年14期
10 孟浩;徐翠平;;虹膜识别算法的研究[J];哈尔滨工程大学学报;2006年03期
，本文编号：2353800
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2353800.html
