ICPR Workshop ETCHB 2010

1,101 views

Published on

Multispectral Palm Image Fusion for Biometric Authentication using Ant Colony Optimization

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,101
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

ICPR Workshop ETCHB 2010

  1. 1. Multispectral Palm Image Fusion for Biometric Authentication using Ant Colony Optimization D. R. Kisku, P. Gupta, J. K. Sing, C. J. Hwang Asansol Engineering College, Asansol – 713305, India Email: drkisku@ieee.org 1
  2. 2. Outline of the Talk: Introduction Intra-modal fusion Advantages over Uni-biometrics systems Palmprint features Advantages of palmprint biometrics Multi-spectral palm image fusion Detection of ROI Wavelet based multi-spectral palm image fusion Gabor representation of fused image Feature selection using ant colony optimization Classification with SVM Experimental results Conclusion 2
  3. 3. Background: Intra-modal Fusion Intra-modal fusion refers to fusion of multiple instances obtained from same modality are linearly combined. Variations of intra-modal fusion: Matching scores obtained from multiple instances of the same modality are fused linearly. Feature level fusion of feature vectors obtained from multiple instances of the same modality Sensor level fusion or image fusion of multiple instances of the same modality 3
  4. 4. Advantages of Intra-modal Fusion: Combining the evidences obtained in different forms from the same or different sources using an effective fusion scheme can significantly improve the overall accuracy of the biometric system. Intra-modal fusion can address the problem of non-universality which often occurs in uni-modal system. Intra-modal systems can provide certain degrees of flexibility. The availability of multiple sources of information can reduce the redundancy in uni-modal system. 4
  5. 5. Palmprint Features: Wrinkles wrinkles are thinner than the principal lines and much more irregular Creases Creases are detailed textures, like the ridges in a fingerprint, all over the palmprint. Creases can only be captured using high resolution cameras. Heart line (a-b) Head line (c-d) Life line (e-f) 5
  6. 6. Advantages of Palmprint Biometrics: Stability User friendliness One may feel comfort to give palmprint images to capturing devices. Acceptability Palmprint is suitable for everyone and it is also non-intrusive as it does not require any personal information of the user. It requires low resolution capturing devices, not like fingerprint devices. Uniqueness Palmprint features do not change over time. 6
  7. 7. Multi-spectral Palm Image Fusion: Biometric palm image fusion at low level [1] refers to a process that fuses multispectral palm images captured by identical or different biometric sensors. The fusion performed at low level produces a fused image in the spatially enhanced form which contains richer, intrinsic and complementary information. [1] D. R. Kisku, J. K. Sing, M. Tistarelli, and P. Gupta, “Multisensor biometric evidence fusion for person authentication using wavelet decomposition and monotonic-decreasing graph,” 7th IEEE International Conference on Advances in 7 Pattern Recognition, pp. 205—208, 2009.
  8. 8. Detection of ROI: Method of ROI (region of interest) detection [2] is employed to reduce the error caused by translation and rotation. This process roughly aligns the palmprint and it does not reduce the effect of palmprint distortion. The main steps of preprocessing - original image, binary image, boundary tracking, building a coordinate system, extracting the central part as a sub-image preprocessed result. [2] D. Zhang, W. K. Kong, J. You, and M. Wong, “On-line palmprint identification,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 25, pp. 1041 – 1050, 2003. 8
  9. 9. 9
  10. 10. Wavelet based Multi-spectral Palm Image Fusion: The wavelet transform [3] provides a multi-resolution decomposition of an image. In wavelet based palm image fusion, decomposition is done with high resolution palmprint images. Decomposition generates a set of low resolution images with wavelet coefficients for each level where the basis functions are generated from one single basis function known as the mother wavelet. The mother wavelet is shifted and scaled to obtain the basis functions. Then, it replaces a low resolution image with a multispectral (MS) band at the same spatial resolution level. Finally, a reverse wavelet transformation is performed to convert the decomposed and set to the original resolution level. The operations of a wavelet fusion scheme are outlined in Fig. 2. [3] T. S. Lee, “Image representation using 2D Gabor wavelets,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 18, pp. 959 – 971, 1996. 10
  11. 11. Contd… The input images are decomposed by a discrete wavelet transform and the wavelet coefficients are selected using a wavelet fusion rule (viz. ‘maximum’ rule) [4] and an inverse discrete wavelet transform is performed to reconstruct the fused image. IDWT DWT First Palm Image Fused Image Fusion of DWT Decompositions Second Palm Image Wavelet Coefficients [4] D. R. Kisku, J. K. Sing, M. Tistarelli, and P. Gupta, “Multisensor biometric evidence fusion for person authentication using wavelet decomposition and monotonic-decreasing graph,” 7th IEEE International Conference on Advances in 11 Pattern Recognition, pp. 205—208, 2009.
  12. 12. Gabor Representation of Fused Image: Fundamentally, 2D Gabor filter [5] can be defined as a linear filter whose impulse response function is the multiplication of harmonic function and Gaussian function in which Gaussian function is modulated by a complex sinusoid. In this regard, the convolution theorem states that the Fourier transform of a Gabor filter's impulse response is the convolution of the Fourier transform of the harmonic function and the Fourier transform of the Gaussian function. Gabor function is a non-orthogonal wavelet and it can be specified by the frequency of the sinusoid and the standard deviations of and. The 2D Gabor wavelet Filter can be defined as 1 (m ) 2 (n) 2 g ( x , y : f , θ ) = exp( − ( 2 + ) cos( 2 π f ( m ))) 2 σ x σ 2y m = x sin θ + y cos θ ; n = x cos θ − y sin θ ; [5] T. S. Lee, “Image representation using 2D Gabor wavelets,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 18, pp. 959 – 971, 1996. 12
  13. 13. Contd… where – f is the frequency of the sinusoidal plane wave along the direction from the x-axis, and specify the Gaussian envelop along x-axis and along y-axis, respectively. This can be used to determine the bandwidth of the Gabor filter. For the sake of experiment, 200 dpi gray scale fused palm image with the size of 40 × 40 has been used. Along with this, 40 spatial frequencies are used with, f = π / 2 (i = 1,2,…,5) i and θ = kπ / 8, (k = 1,2,...,8) . 13
  14. 14. Ant Colony Optimization: Concepts 14
  15. 15. ACO Background: Ants [6] navigate from nest to food source Shortest path is discovered via pheromone trails each ant moves at random pheromone is deposited on path ants detect lead ant’s path, inclined to follow more pheromone on path increases probability of path being followed [6] M. Dorigo, L. M. Gambardella, M. Birattari, A. Martinoli, R. Poli, and T. Stützle, “Ant colony optimization and swarm intelligence,” 5th International Workshop ANTS, LNCS 4150, Springer Verlag, 2006. 15
  16. 16. ACO Algorithm: Starting node is selected randomly Path selected at random based on amount of “trail” present on possible paths from starting node higher probability for paths with more “trail” Ant reaches next node, selects next path Continues until reaches starting node Finished “tour” is a solution A completed tour is analyzed for optimality “Trail” amount adjusted to favor better solutions better solutions receive more trail worse solutions receive less trail higher probability of ant selecting path that is part of a better-performing tour New cycle is performed Repeated until most ants select the same tour on every cycle (convergence to solution) 16
  17. 17. Features Selection using Ant Colony Optimization: An ant chooses the new feature point according to the following β  arg max { Pil H } if q ≤ q 0  l∈ N i k il j =  S otherwise  A random variable selected according to the following probabilistic rule β  P ij H ij  if j ∈ N k  ∑ P il H il β i S =  l ∈ N ik  0 otherwise  17
  18. 18. Contd… Global pheromone update Pij = (1 − σ ) Pij + σ ∆ p bs ij  bs 1/ L  if (i, j) belongs to the best tour ∆pij =  bs 0 otherwise   18
  19. 19. Contd… Local pheromone update Pij = (1 − γ ) Pij + γ p 0 19
  20. 20. Classification with SVM: SVM [7] is known as statistical learning theory which can be used for classification of test samples with respect to training samples. SVMs are built based on the principle of structural risk minimization. The aim is to minimize the upper bound on expected or actual risk which is defined as 1 R (α ) = ∫ z − f ( x , α ) dP ( x , z ) 2 where α is a set of parameters which can be used to define a trained machine z is a class label associated with a training sample x f(x, α) is a function which can be used to mapping training sample to class labels P(x, z) is the unknown probability distribution associating a class label with each training sample. [7] C. J. C. Burges, “A tutorial on support vector machines for pattern recognition,” Data Mining and Knowledge Discovery, vol. 2, no. 2, pp. 121–167, 1998. 20
  21. 21. Contd… Let l denote the number of training samples and choose some η such that 0 ≤ η ≤ 1. On expected risks with the probability 1 - η, the following bound holds h(log( 2l / h) + 1) − log(η / 4) R (α ) ≤ Remp (α ) + l where h is a non-negative integer called Vapnik Charvonenkis (VC) dimension [14] and is a measure of the complexity of the given decision function. The term in R.H.S. is known as VC bound. Risk can be minimized by minimizing the empirical risk as well as VC dimension. 21
  22. 22. Contd… 22
  23. 23. Contd… To separate a given training sample, an optimal hyperplane is chosen from a set of hyperplane. This optimal hyperplane minimizes the VC confidence that provides best generalization capabilities. The optimal hyperplane is used to minimize the sum of the distances to the closest positive and negative training samples. This sum is known as the margin of the separating hyperplane. It can be shown that the optimal hyperplane w—x + b = 0 is obtained by minimizing ||w||2 subject to a set of constraints. Classifiers K ( xi , x j ) = x i • x j K ( xi , x j ) = e − γ || xi − x j || 2 | xT ωk | d C ( x, ω k ) = | x || ω k | 23
  24. 24. Database and Experimental Protocol: Database: CASIA [8] Multispectral Palm Images 3600 Palm Image and 100 Subjects Palm images captured in two different sessions In each session 3 different sets of images captured Each set contains 6 images Between two sets, a certain degree of posture variations is allowed. Experimental protocol: Database is divided into 3 disjoints sets First set (training set) contains 1985 palm images Second set (evaluation set) contains 966 palm images Third set (query set) contains 649 palm images The training set is used to build client models The evaluation set is used to obtain the client and imposter scores for verification thresholds The query set of palm images are used to obtain the verification rates. [8] Y. Hao, Z. Sun, T. Tan, and C. Ren, “Multi spectral palm image fusion for accurate contact free palmprint recognition,” 15th International Conference on Image Processing, pp. 281 – 284, 2008. 24
  25. 25. Experimental Results: Table1. Verification Performance determined on the CASIA MultiSpectral Palm Database CLASSIFIER KERNEL KERNEL EVALUATION SET QUERY SET FUNCTION PARAMETER EER TE FA FR TE NC -------- -------- 6.19% 12.38 4.51% 6.27% 10.78% % SVM Linear -------- 5.02% 10.04 3.09% 5.11% 8.2% % RBF γ = 0.015 3.97% 7.94% 2.21% 4.04% 6.25% 25
  26. 26. Comparative Study: Table 2. Comparison Table Method Fusion Rule Database Classifier EER (%) Method – I [9] CT CASIA Hamming distance 0.5 Method – II [9] SIDWT CASIA Hamming distance 0.58 Proposed Method Haar wavelet with CASIA SVM with RBF 3.97 maximum fusion (Evaluation set) rule SVM with RBF (Query 3.125 set) [9] Y. Hao, Z. Sun, T. Tan, and C. Ren, “Multi spectral palm image fusion for accurate contact free palmprint recognition,” 15th International Conference on Image Processing, pp. 281 – 284, 2008. 26
  27. 27. Conclusion: An efficient palmprint authentication system has presented by fusion of multispectral palm images. Multispectral palm images are fused at low level by wavelet transform and decomposition. This fused palm image is further represented by Gabor wavelet transform to capture the minimal intra-class diversity of the same instances and maximized the inter-class differences between the different subjects in terms of neighborhood pixel intensity changes. Gabor palm responses contain high dimensionality features and due to this high dimensionality, ant colony optimization (ACO) algorithm is applied to choose a set of distinct features. Finally, two different classifiers are used, namely, normalized correlation and SVM with linear and RBF kernels. To measure the efficacy and robustness of the proposed system, CASIA multispectral palm database is used. The results are found to be encouraging. 27
  28. 28. Questions/Comments, if any ! 28

×