基于树形结构小波变换的虹膜纹理分类识别研究
|
摘要
随着网络技术和信息技术的高速发展，人们的安全意识越来越高，同时对安全性提出了更高的要求。目前，常规的安全技术已经不能满足当前的要求，于是人们把目光转向生物识别领域。虹膜具有可靠性、唯一性、不可伪造性及不可侵犯性等优点，从而虹膜识别应运而生。虹膜识别技术能广泛使用在一些安全领域以及海关，它提供了比其它的如指纹，人脸识别等人类特征识别技术更有效的安全性。虹膜结构复杂，1：1035的高防伪几率足以用来作为生物签名。这就意味着在识别过程中不可能找到两个完全相同的虹膜，因此定义一种适合于从人眼图像中提取虹膜信息的表示方法就显得非常重要。
一般情况下，就整幅虹膜原图像来看，所有的虹膜原图像大致上都属于一类图像。但是当仅仅考虑虹膜纹理部分的时候，不同人的虹膜纹理是不一样的，这也是虹膜识别之所以能完成人身份验证的前提。因此我们首先将虹膜图像中的虹膜环带部分分离出来，并进行规范化。虹膜的提取采用了以下方法：通过Canny边缘检测算子对图像进行边缘检测；使用法线算法进行瞳孔中心和虹膜内缘半径的初步估计；在近似认为虹膜内外缘为同心圆的情况下，使用Hough变换方法进行中心定位出虹膜纹理区域，然后通过坐标系变换和双线性插值将虹膜环带部分展开成矩形。
对于提取出来的虹膜纹理图像，如果能够找到一种有效提取虹膜纹理部分特征的方法就能够完成虹膜纹理图像的分类。小波变换是一个时间和
频率的局域变换，因而能有效的从信号中提取信息，通过伸缩和平移等运算功能对函数或信号进行多尺度细化分析，解决了Fourier变换不能解决的许多困难问题，从而小波变化被誉为“数学显微镜”。在以前采用简单的分解虹膜纹理图像低频部分来提取虹膜纹理特征的方法中，很大可能会丢失掉了中频，高频部分的特征信息。所以提取出来的特征数就相对较少，不利于后期进行有效的分类识别。也有人尝试是对所有子带进行分解的完全树形结构小波分解的方法来提取特征。但是，由于每次都要对产生的每个子带进行分解，虽说能提取出虹膜纹理图像的大部分特征信息。但是这样势必增加计算量，影响后期分类识别的速度。所以应该采用一种评价手段，对每次分解出来的子带有选择地进行分解。这样既利于提取出大部分的虹膜纹理信息，又避免了全部树形分解的大量计算。进而提出一种虹膜纹理图像分类算法。
由于每一层小波分解后所产生的待选择分解子带数目不定，所以在评价过程中就得看是全局评价还是局部评价。例如经过第一层分解后的有4个子带有待被选择，第二次分解以后最大可能有16个子带有待被选择等等。首先来说一下局部评价：在分解过程中，仅仅在每一个子带所产生的四个子带中选择被分解子带。全局评价：在分解过程中，在每次分解后的所有子带中进行评价选择能量最大（包含重要的纹理信息）的子带进行下一步分解。很明显我们要用到的是全局评价，因为由评价条件产生的每一次被分解的子带数目不定。这样才能实现树形小波变换的自适应提取重要的纹理信息的特性。全局评价的小波变换算法实现起来比局部评价的树形
小波分解算法难度要大，因为它不能通过简单的递归方法来实现，而是在每次产生的数目不确定的子带中进行评价，从而选择出含重要纹理信息的子带进行下一步分解。根据试验结果看来，被选择作进一步分解的确实不仅仅是图像的低频部分，而是大部分集中在中频带。
为了有利于后期的识别，我们把虹膜特征表示为两部分：第一部分为主频率通道。我们采用1，2，3，4这几个整数来分别表示每次小波分解后的LH，HH，HL，LL子带。这样分解完以后，以评价标准得到一系列由整数表示的分解路径。大量实验验证，对于来自同一个人得虹膜规范化后纹理图像采用树形结构的小波分解路径绝大多数是一样的。对分解路径树按由叶节点到根节点的顺序得到的路径称为一条主频率通道。虹膜纹理特征第二部分就是每一条主频率通道上节点的能量和所组成的向量。以此能量向量的大小我们采用了前五个主频率通道。这两部分都反映了纹理图像的纹理特征，前五个主频率通道注意表示该虹膜纹理特征。
在虹膜识别阶段，首先对来自同一个人得m幅虹膜纹理进行预处理，树形结构小波变换得到它们的主频率通道，对相同主频率通道的能量特征向量进行平均，得到该类虹膜纹理的模板特征。对于一幅未知的虹膜原图像，经过预处理后对其进行同样树形结构的小波变换。得到表征该虹膜纹理特征的主频率通道和对应该主频率通道的能量特征向量。在数据库中找出具有相同主频率通道的虹膜模板特征。在同一通道中比较它们的能量特征向量，如果差异值小于某给定的阈值，就把该纹理归为此类虹膜纹理，从而完成虹膜图像的分类识别。
实验验证了该方法在小样本空间下有效性，今后我们工作应在较大的样本空间上继续进行理论证明和实验验证。
Following the Internet and information technology development, the security consciousness of the people became more and more high and the claim of security became more and more high. At present, the routine security technology doesn’t meet the requirement of security quality of the people, then the people turn to biology recognition technology to meet the requirement. For iris has some good advantage, such as reliability, unique, and it can’t forge and it can’t infringe, then iris recognition is put forward. Iris recognition can be widely used in security and customs, and it provides superiority security than other human feature recognition such as fingerprint, face and so on. The iris is complex enough to be used as a biometric signature with imposter odds ranging as high as 1 in 10. It means that the improbability of finding two people with identical iris pattern for identification, thus it is important to define a representation that is well adapted to extract the iris information context from iris texture images.
In general, all source iris images are congener images, but when it comes to the iris area, Iris of different persons is different. This is the crucial precondition of iris identification. So we should first detect the iris area from the source iris image. In the course of this we adopt the following approach: detect the iris edge by Canny operator; estimate the center of pupil and size of inner edge of iris; and use the Hough transform to detect the inner and outer edges of iris on the consideration that their have the same center; finally, use Normalization Line Algorithm to transform the iris annular area to rectangular area.
For the normalized iris texture images, if we find a effective way to extract their characters, then we can classify the iris images. The wavelet transform is a transform about time and frequency, so the information can be extracted by the wavelet transform. The mother wavelet has the translation and scale parameters, which change respectively. The sign can be analyzed into Multiresolution by the wavelet transform, and it can solve many difficult
question which can’t be solved by the Fourier transform. For the wavelet transform have the translation and scale propriety, then the wavelet transform is called “microscope”. In the preceding method based on pyramid-structured wavelet transform, for the reason that the decomposition is implied recursively to the low frequency region, these approaches is very likely to loss much information in middle or high frequency region. As a result, the information extracted is sufficient enough for the latter identification. Some researchers attempted to decompose all this frequency regions to extract more information of the texture images. But this added the complication computation. In order to obtain the important information of the iris texture image and avoid a full decomposition. We consider a criterion to decide whether a further decomposition is needed for a particular output. Based on this realization, we give a newly approach of iris identification.
As above statement, for the uncertainty of the number of subimages after decomposition every time, we should consider all-around or local evaluation. For example, there are four subimages to be chosen after the first decomposition, but the number of these subimages becomes sixteen after second decomposition, etc. the local evaluation: in this approach, we only choose the subimage to further decomposition from the four subimages of some subimage after decomposition every time. While in the all-around evaluation, the subimage to further decomposition is chosen from all subimages after decomposition every time. The subimage contains more texture information by the criterion. Obviously, we should adopt the latter approach; only in this way can we embody the adaptability of tree-structured wavelet transform. The decomposition of all-around evaluation is not easy to implement for uncertainty of the number of subimages and it can’t be performed by recursive way. From our experime
一般情况下，就整幅虹膜原图像来看，所有的虹膜原图像大致上都属于一类图像。但是当仅仅考虑虹膜纹理部分的时候，不同人的虹膜纹理是不一样的，这也是虹膜识别之所以能完成人身份验证的前提。因此我们首先将虹膜图像中的虹膜环带部分分离出来，并进行规范化。虹膜的提取采用了以下方法：通过Canny边缘检测算子对图像进行边缘检测；使用法线算法进行瞳孔中心和虹膜内缘半径的初步估计；在近似认为虹膜内外缘为同心圆的情况下，使用Hough变换方法进行中心定位出虹膜纹理区域，然后通过坐标系变换和双线性插值将虹膜环带部分展开成矩形。
对于提取出来的虹膜纹理图像，如果能够找到一种有效提取虹膜纹理部分特征的方法就能够完成虹膜纹理图像的分类。小波变换是一个时间和
频率的局域变换，因而能有效的从信号中提取信息，通过伸缩和平移等运算功能对函数或信号进行多尺度细化分析，解决了Fourier变换不能解决的许多困难问题，从而小波变化被誉为“数学显微镜”。在以前采用简单的分解虹膜纹理图像低频部分来提取虹膜纹理特征的方法中，很大可能会丢失掉了中频，高频部分的特征信息。所以提取出来的特征数就相对较少，不利于后期进行有效的分类识别。也有人尝试是对所有子带进行分解的完全树形结构小波分解的方法来提取特征。但是，由于每次都要对产生的每个子带进行分解，虽说能提取出虹膜纹理图像的大部分特征信息。但是这样势必增加计算量，影响后期分类识别的速度。所以应该采用一种评价手段，对每次分解出来的子带有选择地进行分解。这样既利于提取出大部分的虹膜纹理信息，又避免了全部树形分解的大量计算。进而提出一种虹膜纹理图像分类算法。
由于每一层小波分解后所产生的待选择分解子带数目不定，所以在评价过程中就得看是全局评价还是局部评价。例如经过第一层分解后的有4个子带有待被选择，第二次分解以后最大可能有16个子带有待被选择等等。首先来说一下局部评价：在分解过程中，仅仅在每一个子带所产生的四个子带中选择被分解子带。全局评价：在分解过程中，在每次分解后的所有子带中进行评价选择能量最大（包含重要的纹理信息）的子带进行下一步分解。很明显我们要用到的是全局评价，因为由评价条件产生的每一次被分解的子带数目不定。这样才能实现树形小波变换的自适应提取重要的纹理信息的特性。全局评价的小波变换算法实现起来比局部评价的树形
小波分解算法难度要大，因为它不能通过简单的递归方法来实现，而是在每次产生的数目不确定的子带中进行评价，从而选择出含重要纹理信息的子带进行下一步分解。根据试验结果看来，被选择作进一步分解的确实不仅仅是图像的低频部分，而是大部分集中在中频带。
为了有利于后期的识别，我们把虹膜特征表示为两部分：第一部分为主频率通道。我们采用1，2，3，4这几个整数来分别表示每次小波分解后的LH，HH，HL，LL子带。这样分解完以后，以评价标准得到一系列由整数表示的分解路径。大量实验验证，对于来自同一个人得虹膜规范化后纹理图像采用树形结构的小波分解路径绝大多数是一样的。对分解路径树按由叶节点到根节点的顺序得到的路径称为一条主频率通道。虹膜纹理特征第二部分就是每一条主频率通道上节点的能量和所组成的向量。以此能量向量的大小我们采用了前五个主频率通道。这两部分都反映了纹理图像的纹理特征，前五个主频率通道注意表示该虹膜纹理特征。
在虹膜识别阶段，首先对来自同一个人得m幅虹膜纹理进行预处理，树形结构小波变换得到它们的主频率通道，对相同主频率通道的能量特征向量进行平均，得到该类虹膜纹理的模板特征。对于一幅未知的虹膜原图像，经过预处理后对其进行同样树形结构的小波变换。得到表征该虹膜纹理特征的主频率通道和对应该主频率通道的能量特征向量。在数据库中找出具有相同主频率通道的虹膜模板特征。在同一通道中比较它们的能量特征向量，如果差异值小于某给定的阈值，就把该纹理归为此类虹膜纹理，从而完成虹膜图像的分类识别。
实验验证了该方法在小样本空间下有效性，今后我们工作应在较大的样本空间上继续进行理论证明和实验验证。
Following the Internet and information technology development, the security consciousness of the people became more and more high and the claim of security became more and more high. At present, the routine security technology doesn’t meet the requirement of security quality of the people, then the people turn to biology recognition technology to meet the requirement. For iris has some good advantage, such as reliability, unique, and it can’t forge and it can’t infringe, then iris recognition is put forward. Iris recognition can be widely used in security and customs, and it provides superiority security than other human feature recognition such as fingerprint, face and so on. The iris is complex enough to be used as a biometric signature with imposter odds ranging as high as 1 in 10. It means that the improbability of finding two people with identical iris pattern for identification, thus it is important to define a representation that is well adapted to extract the iris information context from iris texture images.
In general, all source iris images are congener images, but when it comes to the iris area, Iris of different persons is different. This is the crucial precondition of iris identification. So we should first detect the iris area from the source iris image. In the course of this we adopt the following approach: detect the iris edge by Canny operator; estimate the center of pupil and size of inner edge of iris; and use the Hough transform to detect the inner and outer edges of iris on the consideration that their have the same center; finally, use Normalization Line Algorithm to transform the iris annular area to rectangular area.
For the normalized iris texture images, if we find a effective way to extract their characters, then we can classify the iris images. The wavelet transform is a transform about time and frequency, so the information can be extracted by the wavelet transform. The mother wavelet has the translation and scale parameters, which change respectively. The sign can be analyzed into Multiresolution by the wavelet transform, and it can solve many difficult
question which can’t be solved by the Fourier transform. For the wavelet transform have the translation and scale propriety, then the wavelet transform is called “microscope”. In the preceding method based on pyramid-structured wavelet transform, for the reason that the decomposition is implied recursively to the low frequency region, these approaches is very likely to loss much information in middle or high frequency region. As a result, the information extracted is sufficient enough for the latter identification. Some researchers attempted to decompose all this frequency regions to extract more information of the texture images. But this added the complication computation. In order to obtain the important information of the iris texture image and avoid a full decomposition. We consider a criterion to decide whether a further decomposition is needed for a particular output. Based on this realization, we give a newly approach of iris identification.
As above statement, for the uncertainty of the number of subimages after decomposition every time, we should consider all-around or local evaluation. For example, there are four subimages to be chosen after the first decomposition, but the number of these subimages becomes sixteen after second decomposition, etc. the local evaluation: in this approach, we only choose the subimage to further decomposition from the four subimages of some subimage after decomposition every time. While in the all-around evaluation, the subimage to further decomposition is chosen from all subimages after decomposition every time. The subimage contains more texture information by the criterion. Obviously, we should adopt the latter approach; only in this way can we embody the adaptability of tree-structured wavelet transform. The decomposition of all-around evaluation is not easy to implement for uncertainty of the number of subimages and it can’t be performed by recursive way. From our experime
引文
[1] Espen Norheim (2000) “A Human Identification System Based on Images of the Iris” Undergraduate project report Queensland University of Technology Brisbane.
[2] L.Flom and A.Safir “Iris Recognition System”. U.S. patent 4641349,1987
[3] J.Daugman “High Confidence Visual Recognition of Persons by A Test of Statistical Independence” IEEE Trans Pattern Anal. Machine Intelligence, 1993, 15 (11): 1148~1161
[4] Wildes,“Automated,Non-invasive Iris Recognition System and Method”, U.S. patent 5751836, 1998.
[5] R.P.Widers, J.C.Asmuth, “A Machine Vision System For Iris Recognition”, Machine Vision applicat. Vol.9, pp1～8. 1996
[6] J.Daugman, “Biometric Personal Identification System Based on Iris Analysis”, U.S. patent 5291560, March 1, 1994
[7] W.W.Boles, “A Security System Based on Human Iris Identification Using Wavelet Transform”, 1997 First International Conference on Knowledge-Based Intelligent Electronic System, May 1997
[8] Wildes R P., “Iris Recognition: An Merging Biometric Technology”, Proceeding of The IEEE, Vol.85, NO.9, pp1348～1363, Sep 1997
[9] Shinyoung Lim, Kwanyong Lee, Okhwan Byeon and Taiyun Kim, “Efficient Iris Recognition through Improvement of Feature Vector and Classifier,” ETRI Journal, Vol. 23, no. 2, pg 61～70, June 2001.
[10] Christel-lo?c TISSE, Lionel MARTIN, Lionel TORRES , Michel ROBERT, “Person identification technique using human iris Recognition”, The 15th International Conference on Vision Interface，May 27-29, 2002, Calgary, Canada
[11] Eric Sung, “ Towards non-cooperative iris recognition systems”, Proceedings of the 7th International Conference on Control, Automation, Robotics and Vision ,2002
[12] 王蕴红,朱勇,谭铁牛，“基于虹膜识别的身份鉴别”，自动化学报，2002 Vol.28 No.1 P.1～10
[13] 谭铁牛,朱勇,王蕴红. "虹膜图像采集装置"实用新型.国家知识产权局专利号: ZL 99217063.X
[14] Boles W, Boashah B，A human identification technique using images of the iris and wavelet transform. IEEE Trans. on Signal Processing, 1998, 46(4):1185～1188
[15] 应忍冬，徐国治，基于小波变换过零检测的虹膜识别技术，上海交通大学学报，2002.3, Vol.36 No.3 p354～p358
[16] 康 浩 ,徐国,治虹膜纹理的相位编码, 上海交通大学学报，May 1999, Vol.33 No.5 p542～p354
[17] 康浩、徐国治. 虹膜纹理的能量编码,电路与系统学报,2000.3 Vol.5 No.1 p444～492
18] 何家峰、廖曙铮、叶虎年、李柱. 虹膜定位. 中国图象图形报, 2000, 5(3): 253~255
[19] Chang T, Kuo C.-C. J. Texture Analysis and Classification with Tree-Structured Wavelet Transform. IEEE Trans. Image Prossesiing, 1993, 2(10): 429~221.
[20] 廖曙铮，叶虎年，一种序列虹膜图像的质量评价方法
[21] 李庆嵘，虹膜定位算法研究，电子科技大学学报，Feb.2002，Vol.31No.1 p7～p9
[22] Yuan-Ning liu, Xing Ming, Sen-Miao Yuan, ZHI-HUI LI,” An Iris Recognition System Including Acquisition and Locating as well as Texture Representation”, ICWAA2003,2003
[23] Yuan-ning liu, Zhi-hui li, Sen-Miao Yuan, TAO XU,” Iris Image Analysis And Reconstruction Based On Zero-Crosing Wavelet Transform”, ICWAA2003, 2003
[24] 严民军，齐翔林，汪云九。“虹膜识别中的一种神经算法的研究”，生物物理学报，第十六卷，第四期，2000.12
[26] 贾云得，《机器视觉》，科学出版社，2000
[27] 章毓晋，《图象分割》，科学出版社，2001
[28] 徐涛. 数值计算方法. 吉林科学技术出版社. 1998.7
[29] K.R.Castleman著，朱志刚等译， 《数字图象处理》，电子工业出版社，1998
[30] W.K.Leow and R.li “Adaptive binning and dissimilarity measure for image retrieval and classification” in Proc.IEEE CVPR 2001,2001.
[31] Tianhorng Chang and C. -C.Jay Kuo “Tree-Structured Wavelet Transform for Textured Image Segmentation” Signal and Image Processing Institute and Department of Electrical Engineering-Systems University Of Southern California, Los Angeles, California 90089-2564 SPIE Vol 1770(1992)/397.
[32] 李志慧. 基于小波过零点的虹膜识别技术的研究. 硕士毕业论文，吉林大学. 2003
[2] L.Flom and A.Safir “Iris Recognition System”. U.S. patent 4641349,1987
[3] J.Daugman “High Confidence Visual Recognition of Persons by A Test of Statistical Independence” IEEE Trans Pattern Anal. Machine Intelligence, 1993, 15 (11): 1148~1161
[4] Wildes,“Automated,Non-invasive Iris Recognition System and Method”, U.S. patent 5751836, 1998.
[5] R.P.Widers, J.C.Asmuth, “A Machine Vision System For Iris Recognition”, Machine Vision applicat. Vol.9, pp1～8. 1996
[6] J.Daugman, “Biometric Personal Identification System Based on Iris Analysis”, U.S. patent 5291560, March 1, 1994
[7] W.W.Boles, “A Security System Based on Human Iris Identification Using Wavelet Transform”, 1997 First International Conference on Knowledge-Based Intelligent Electronic System, May 1997
[8] Wildes R P., “Iris Recognition: An Merging Biometric Technology”, Proceeding of The IEEE, Vol.85, NO.9, pp1348～1363, Sep 1997
[9] Shinyoung Lim, Kwanyong Lee, Okhwan Byeon and Taiyun Kim, “Efficient Iris Recognition through Improvement of Feature Vector and Classifier,” ETRI Journal, Vol. 23, no. 2, pg 61～70, June 2001.
[10] Christel-lo?c TISSE, Lionel MARTIN, Lionel TORRES , Michel ROBERT, “Person identification technique using human iris Recognition”, The 15th International Conference on Vision Interface，May 27-29, 2002, Calgary, Canada
[11] Eric Sung, “ Towards non-cooperative iris recognition systems”, Proceedings of the 7th International Conference on Control, Automation, Robotics and Vision ,2002
[12] 王蕴红,朱勇,谭铁牛，“基于虹膜识别的身份鉴别”，自动化学报，2002 Vol.28 No.1 P.1～10
[13] 谭铁牛,朱勇,王蕴红. "虹膜图像采集装置"实用新型.国家知识产权局专利号: ZL 99217063.X
[14] Boles W, Boashah B，A human identification technique using images of the iris and wavelet transform. IEEE Trans. on Signal Processing, 1998, 46(4):1185～1188
[15] 应忍冬，徐国治，基于小波变换过零检测的虹膜识别技术，上海交通大学学报，2002.3, Vol.36 No.3 p354～p358
[16] 康 浩 ,徐国,治虹膜纹理的相位编码, 上海交通大学学报，May 1999, Vol.33 No.5 p542～p354
[17] 康浩、徐国治. 虹膜纹理的能量编码,电路与系统学报,2000.3 Vol.5 No.1 p444～492
18] 何家峰、廖曙铮、叶虎年、李柱. 虹膜定位. 中国图象图形报, 2000, 5(3): 253~255
[19] Chang T, Kuo C.-C. J. Texture Analysis and Classification with Tree-Structured Wavelet Transform. IEEE Trans. Image Prossesiing, 1993, 2(10): 429~221.
[20] 廖曙铮，叶虎年，一种序列虹膜图像的质量评价方法
[21] 李庆嵘，虹膜定位算法研究，电子科技大学学报，Feb.2002，Vol.31No.1 p7～p9
[22] Yuan-Ning liu, Xing Ming, Sen-Miao Yuan, ZHI-HUI LI,” An Iris Recognition System Including Acquisition and Locating as well as Texture Representation”, ICWAA2003,2003
[23] Yuan-ning liu, Zhi-hui li, Sen-Miao Yuan, TAO XU,” Iris Image Analysis And Reconstruction Based On Zero-Crosing Wavelet Transform”, ICWAA2003, 2003
[24] 严民军，齐翔林，汪云九。“虹膜识别中的一种神经算法的研究”，生物物理学报，第十六卷，第四期，2000.12
[26] 贾云得，《机器视觉》，科学出版社，2000
[27] 章毓晋，《图象分割》，科学出版社，2001
[28] 徐涛. 数值计算方法. 吉林科学技术出版社. 1998.7
[29] K.R.Castleman著，朱志刚等译， 《数字图象处理》，电子工业出版社，1998
[30] W.K.Leow and R.li “Adaptive binning and dissimilarity measure for image retrieval and classification” in Proc.IEEE CVPR 2001,2001.
[31] Tianhorng Chang and C. -C.Jay Kuo “Tree-Structured Wavelet Transform for Textured Image Segmentation” Signal and Image Processing Institute and Department of Electrical Engineering-Systems University Of Southern California, Los Angeles, California 90089-2564 SPIE Vol 1770(1992)/397.
[32] 李志慧. 基于小波过零点的虹膜识别技术的研究. 硕士毕业论文，吉林大学. 2003