A Performance Evaluation of Shape and Texture Based Methods for Vein Recognition

Please download to get full document.

View again

of 3
18 views
PDF
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Document Description
A Performance Evaluation of Shape and Texture Based Methods for Vein Recognition
Document Share
Document Tags
Document Transcript
  A performance Evaluation of Shape and Texture based methods for Vein Recognition *Zhongli Wang, **Baochang Zhang, Weiping Chen, Yongsheng Gao *Computer Science and Engineering, Beijing Institute of Technology,P.R.China **Institute for Integrated and Intelligent System, Griffith University, Australia  zlwang@mae.cuhk.edu.hk, bczhang@jdl.ac.cn Abstract This paper gives fair comparisons of shape and texture based methods for vein recognition. The shape of the back of hand contains information that is capable of authenticating the identity of an individual. In this paper, two kinds of shape matching method are used, which are based on Hausdorff distance and Line Edge  Mapping(LEM) methods. The vein image also contains valuable texture information, and Gabor wavelet is exploited to extract the discriminative feature. In order to evaluate the system performance, a dataset of 100  persons of different ages above 16 and of different  gender, each has 5 images per person is used.  Experimental results show that Hausdorff, LEM and Gabor based methods achieved 58%, 66%, 80% individually. 1. Introduction Biometric is the science of identifying a person using the physiological or behavioral features. Recently, vein  pattern has attracted increasing attention. The physical shape of the subcutaneous vascular tree of the back of the hand contains information that is capable of authenticating the identity of an individual[1,2]. A vein  pattern is the vast network of blood vessels underneath a  person’s skin. It is like fingerprint, which is widely used in person identification. However, compared with fingerprint, vein patterns are mostly invisible to human eye, therefore, they are more stable biometric features for person identity verification. A typical vein pattern biometric system consists of five individual processing stages: image acquisition, image enhancement, vein pattern segmentation, feature extraction and matching. The database we used is from University of Tennessee, one can refer to[2].As for the similarity between vein and fingerprint, the researchers can get good results by using fingerprint recognition method, such as Gabor based method [5]. In this paper, shape and texture based methods are exploited for the  performance evaluation. For shape based methods, Hausdoff and LEM are used as the distance measure, while for the texture based method, Gabor magnitude feature is combined with nearest neighbour classifier using the Euclidian similarity measure. The rest part of paper is organized as follows. In section 2, image enhancement and feature matching methods for the shape based vein recognition are briefly introduced. In section 3, Gabor based method is  proposed for vein recognition. In section 4, comparative experiment is conducted on a database of 500 image for 100 persons. Conclusion and future work appear in section 5. 2. Image enhancement and shape matching for vein recognition The key step for shape based vein recognition is  binary preprocessing for the grey-level image. As for the vein vary across the image, global thresholding techniques alone do not provide satisfactory results. This  paper chooses different threshold value for every pixel in the image based on the local neighbors. 2.1. Binary image for vein pattern The quality of the vein image is not well for global  binary method because the vein database was acquired using the low-cost device. Due to the fact that the gray-level intensity values of the vein vary slowly across the image, global thresholding techniques alone do not  provide the satisfactory results. To obtain a better representation of the shape of the vein pattern, we mainly used local thresh holding method based on local mean value based on a given threshold  g  T  . , 1(,).  MN ijij  Iij MN   µ   = ∑∑  (1) 2008 Congress on Image and Signal Processing 978-0-7695-3119-9/08 $25.00 © 2008 IEEEDOI 10.1109/CISP.2008.106659   2008 Congress on Image and Signal Processing 978-0-7695-3119-9/08 $25.00 © 2008 IEEEDOI 10.1109/CISP.2008.106659   2008 Congress on Image and Signal Processing 978-0-7695-3119-9/08 $25.00 © 2008 IEEEDOI 10.1109/CISP.2008.106659 Authorized licensed use limited to: GRIFFITH UNIVERSITY. Downloaded on April 28, 2009 at 02:25 from IEEE Xplore. Restrictions apply.    where  M   and  N   are indices of local rectangular region. , 1, if (,)()(,)0, i j g   I i j T  I i jotherwise  µ  >= −  =    (2) In this paper, 5  g  T   =  is used as the common offset for all pixels, which are chosen on trial-and-error basis, and local region is set as 1313 × . One sample in Fig.1 shows the vein pattern has been successfully segmented from the srcinal image after applying the local thresholding and thinning algorithms. a) b) Figure 1. a) is srcinal vein image, b) is binary image for a). 2.2. Hausdorff based vein recognition In this section, the hausdorff based shape matching method is briefly described. The hausdorff distance is defined as follows: (,)max{min{(,)}} a Ab B h A B d a b ∈∈ =  (3) where a and b are points of sets A and B respectively, and (,) d a b  is any metric between these points; for simplicity, we will take (,) d a b  as the Euclidian distance  between a and b. After binary vein images are calculated using method in section 2, Eq. 3 is used for the identity classification. Y. Gao, et al. proposed another kind of the hausdorff distance, the so-called Line Edge Map(LEM), for face recognition. This paper also uses it for vein recognition. Details about LEM can refer to [3]. 3. Gabor feature based vein recognition Gabor wavelet models quite well the receptive field  profiles of cortical sim ple cells, therefore, Gabor feature can capture the salient visual properties such as the spatial localization, orientation selectivity, and spatial frequency characteristic[4,5]. The Gabor wavelets (kernels, filters) can be defined as follows: 2222,, (||||||||/2)22/2,, ψ ()(||||/)  u v u v iu v u v  e e e σ   σ   σ   − −   = −   kzkz zk   i  (4) where , cossin v uu vv u k k  Φ   =   Φ   k  , max2 2 v v  f k  + = , 8 u  u π  Φ = , v  is the scale(frequency), and u is the orientation with max 2/2  f   π  =  and (,) T   x y = z . We used 5 scales and 8 orientations for vein recognition. It is well-known that Gabor wavelets can be used to enhance the features in certain scales and orientations, and have been widely used in image processing and object recognition. Figure 2. A example divided into 4 rectangular regions Let , () u v G Z   denotes the Gabor magnitude feature of an image, where u  and v  are the orientation and scale of the kernel, respectively. The feature extraction is  based on the multi-region (see Fig.2) mean Gabor magnitude feature, represented by  ____ ,, u v l  G  µ  as  ____ ,,, 1() i l  u v l  u v i Z R G G Z  N   µ  ∈ =  ∑  (5) where 1,2,..., l L = is the region index, and    L  is the number of sub-region.  N   represents the total number of  pixels in the region l   R . Details about the method can refer to [5]. In the classification phase, the nearest neighbor classifier is used with the Euclidian distance as the similarity measure. In the experiment, 64  L  = . 4. Experiments In this section, a database of 500 images for 100  persons is used to compare the performance of the shape and texture based methods. In the database, each person has 5 images, one is used as target set, others are in query set. Two persons’ samples are shown in Fig.3. In the experiment, the vein images are normalized to the size of 100x100. 660   660   660 Authorized licensed use limited to: GRIFFITH UNIVERSITY. Downloaded on April 28, 2009 at 02:25 from IEEE Xplore. Restrictions apply.    Figure 3. amples of Vein, the first row is labeled as 00001, the second row is labeled as 00002. As shown in Table.1, Gabor feature get the best  performance, and line edge map method achieves much  better performance than point-based hausdorff method. It is interesting that Gabor feature can get much better  performance than traditional shape based methods, the reason for which is that Gabor wavelet is robust to variation on gray-level images. In addition, region-based method can reserve the structure information, which is very important for vein recognition. We should mention that the normalized images are not well alignment, which are one of main problem for vein recognition system. 586680 5055606570758085    R  e  c  o  g  n   i   t   i  o  n   R  a   t  e  s Hausdorff LEMGabor  Fig. 4. Recognition rates for shape and texture based vein recognition methods 5. Conclusion and future work This paper gives fair performance comparison for vein recognition, which is one of most secure  biometric recognition methods. A database of 500 images for 100 persons is used to evaluate three methods, point-based hausdorff, line edge map, and Gabor based methods. The Gabor feature gets the  best performance with 80% recognition rates, line edge map gets 66%, which is much than point-based hausdorff based method. In the future work, we will make further investigation about the image enhancement and feature extraction. Acknowledgment We would like to express our sincere appreciation to Ahmed M. Badawi from University of Tennessee, who share the valuable database to us. 6. References [1] Ahmed M. Badawi, “Hand Vein Biometric Verification Pr ototype: A Testing Performance and Patterns Similar ity,”  In Proceedings of the International Conference on I mage  Processing, Computer Vision, and  Pattern Recognition , 2006, pp.3-9 [2] Lingyu Wang, Graham Leedham, Siu-Yeung Cho, “Minutiae feature analysis for infrared hand vein pattern  biometrics,”  Pattern Recognition , 2007. [3] Y. Gao and M. Leung, “Face Recognition Using Line Edge Map,”  IE  EE Transactions on Pattern Analysis and  M achine Intelligence , Vol. 24, No.6, 2002, pp.764-809 [4] D. Gabor, “Theory of communication,”  Journal o f the  Institution of Electrical Engineers , vol. 93, part III, no. 26, 1946, pp. 429-457 [5] A. K. Jain, S. Prabhakar and L. Hong, "A Multichannel Approach to Fingerprint Classification",  IEEE Transactions on Pattern Analysis and Machine  Intelligence , Vol.21, No.4, 1999, pp. 348-359 661   661   661 Authorized licensed use limited to: GRIFFITH UNIVERSITY. Downloaded on April 28, 2009 at 02:25 from IEEE Xplore. Restrictions apply.
Similar documents
View more...
Search Related
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks