基于广义高斯分布FDCT_Wrap和FSVM的虹膜识别

doi:10.11959/j.issn.1000-0801.2016200

摘要/Abstract

摘要：

为了提高虹膜识别的准确率，提出了一种改进曲波变换的虹膜识别算法。首先对预处理后的虹膜图像进行Wrapping算法的快速离散曲波变换，提取不同尺度和不同方向的曲波子带系数矩阵的均值、方差和能量，然后利用广义高斯分布估算各子带的权值，为分类能力较强的特征向量赋予较大权值，构成虹膜图像的特征向量。最后采用模糊支持向量机和二叉决策树相结合的分类器进行匹配识别。采用UBIRIS和CASIA虹膜数据库对算法性能进行测试。实验结果表明，该算法能更好地提高虹膜识别准确率和效率，具有可行性。

关键词: 广义高斯分布, 虹膜识别, 曲波变换, 模糊支持向量机, 二叉决策树

Abstract:

In order to improve the accuracy rate of iris recognition,an improved curvelet transform algorithm for iris recognition was proposed.Firstly,the iris image was decomposed with fast discrete curvelet transform by wrapping algorithm.Mean variance and energy of curvelet sub-band coefficients in different scales and different orientations were extracted.The weights of sub-bands were estimated by generalized Gaussian distribution.The feature vectors with stronger classification ability had large weight,which were calculated to constitute feature vectors of iris image.Finally,feature vectors were matched and recognized by classifier combined with fuzzy support vector machine and binary decision tree.The algorithm performances were tested with UBIRIS and CASIA iris database.Simulation results show that the proposed algorithm has higher recognition accuracy rate and efficiency.It is feasibility.

Key words: generalized Gaussian distribution, iris recognition, curvelet transform, fuzzy support vector machine, binary decision tree

何振红. 基于广义高斯分布FDCT_Wrap和FSVM的虹膜识别[J]. 电信科学, 2016, 32(7): 126-131.

Zhenhong HE. Iris recognition based on generalized Gaussian distribution FDCT_Wrap and FSVM[J]. Telecommunications Science, 2016, 32(7): 126-131.

图/表 10

图1

图2

图3

图4

表1

表2

表3

表4

图5

图6

参考文献 12

[1]	孙冬，周俊，魏勇 . 基于Haar小波的虹膜特征提取算法[J]. 科学技术与工程， 2014,1(14): 81-85. SUN D , ZHOU J , WEI Y . Iris feature extraction based on Haar wavelet[J]. Science Technology and Engineering, 2014,1(14): 81-85.
[2]	FAROUK R M . Iris recognition based on elastic graph matching and Gabor wavelets[J]. Computer Vision and Image Understanding, 2011,115(8): 1239-1244.
[3]	RAJBHOJ S M , MANE P B . Haar wavelet approach of iris texture extraction for personal recognition[J]. International Journal of Innovative Technology and Exploring Engineering, 2013,3(2): 22-25.
[4]	KARTHIKEYAN V . Novel algorithm for iris biometrics using fast wavelet transforms[J]. International Journal of Engineering Sciences and Research Technology, 2014,3(3): 1481-1484.
	RAHULKAR A D , JADHAV D V , HOLAMBE R S . Fast discrete curvelet transform based anisotropic iris coding and recognition using k-out-of-n: a fused post-classifier[J]. Machine Vision and Applications, 2012,23(6): 1115-1127.
[6]	DHAGEA S S , HEGDEA S S , MANIKANTANA K , et al. DWT-based feature extraction and radon transform based contrast enhancement for improved iris recognition[J]. Procedia Computer Science, 2015,1(45): 256-265.
[7]	ABDULLAH H N , ABDULLAH A A . Iris recognition using wavelet transform and artificial neural networks[J]. Engineering＆ Technology Journal, 2015,33(6): 877-888.
[8]	CANDES E J , DEMANET L , DONOHO D L , et al. Fast discrete curvelet transforms[J]. Multiscale Modeling ＆ Simulation, 2006,5(3): 861-899.
[9]	VAPNIKV N . An overview of statistical learning theory[J]. IEEE Transactions on Neural Networks, 1999,10(5): 988-999.
[10]	LIN C F , WANG S D . Fuzzy support vector machines[J]. IEEE Transactions on Neural Networks, 2002,13(2): 464-471.
[11]	PROENCA H , ALEXANDRE LA . UBIRIS: a noisy iris image database[J]. Lecture Notes in Computer Science, 2005(3617): 970-977.
[12]	CASIA-Iris V2[EB/OL]. (2007-12-20)[2015-12-14]. .

方法	特征向量	正确识别率	错误识别率
未特征加权	低频均值方差	91.15%	8.85%
	高频能量	80.94%	19.06%
	组合向量	93.97%	6.03%
特征加权	低频均值方差	93.23%	6.77%
	高频能量	83.48%	16.52%
	本文提出组合向量	98.16%	1.84%

方法	特征向量	正确识别率	错误识别率
未特征加权	低频均值方差	91.79%	8.21%
	高频能量	82.87%	17.13%
	组合向量	95.22%	4.78%
特征加权	低频均值方差	93.86%	6.14%
	高频能量	84.57%	15.43%
	本文提出组合向量	98.83%	1.17%

组别	特征提取方法	匹配方法	错误识别率	错误拒绝率	正确识别率
1	Gabor小波	HD	4.32%	3.49%	92.94%
	Haar小波		4.71%	2.83%	92.14%
	curvelet变换		2.82%	2.55%	94.28%
	改进curvelet变换		2.47%	2.38%	95.16%
2	Gabor小波	SVM	3.13%	1.91%	94.85%
	Haar小波		4.47%	2.86%	92.72%
	curvelet变换		2.38%	2.04%	95.18%
	改进curvelet变换		1.92%	1.76%	96.37%
3	Gabor小波	二叉决策树FSVM	1.86%	1.17%	96.09%
	Haar小波		2.89%	1.56%	94.53%
	curvelet变换		1.26%	0.78%	97.95%
	改进curvelet变换		1.01%	0.62%	98.16%

组别	特征提取方法	匹配方法	错误识别率	错误拒绝率	正确识别率
1	Gabor小波	HD	4.28%	3.25%	93.13%
	Haar小波		4.66%	2.71%	92.63%
	curvelet变换		2.56%	2.38%	94.74%
	改进curvelet变换		2.13%	2.06%	95.81%
2	Gabor小波	SVM	2.95%	1.74%	95.28%
	Haar小波		4.28%	2.65%	93.07%
	curvelet变换		2.26%	1.92%	95.61%
	改进curvelet变换		1.81%	1.63%	96.56%
3	Gabor小波	二叉决策树FSVM	1.78%	1.09%	96.25%
	Haar小波		2.71%	1.37%	94.68%
	curvelet变换		1.17%	0.61%	98.22%
	改进curvelet变换		0.91%	0.47%	98.83%