RAMAKRISHNA LANKA MSEE, UTA ADVISING PROFESSOR: DR. K. R. RAO

Transcription

1

2 RAMAKRISHNA LANKA MSEE, UTA ADVISING PROFESSOR: DR. K. R. RAO

3 Personal identity management 3 Exponential growth of population and migration of people is a challenge to person management. Risk of identity theft. Crucial societal applications are based on personal identity. Personal attributes define the unique identity.

4 Personal identity management 4 Aishwarya Rai Figure 1: Traditional schemes to validate individuals [1] Figure 2: Biometrics to validate individuals [1]

5 Need for biometrics 5 Traditional person recognition relies on surrogate representations of identity. Knowledge based person recognition mechanisms like PINs and passwords are not reliable. Biometric recognition, or simply biometrics, offers a natural and more reliable solution. Biometric identifiers are inherent.

6 Biometric Systems 6 The term biometrics is derived from the Greek words bio (life) and metric (to measure). Biometrics is the measurable biological and behavioral trait unique to a person. Examples: fingerprints, face, iris, voice, gait, or the Deoxyribonucleic acid (DNA).

7 Biometric Systems 7 There are 2 phases to a biometric system: Enrollment Recognition The biometric system consists of 4 basis components: Sensor Feature extractor Database Matcher

8 Biometric Systems 8 Figure 3: Block diagram of a biometric system [1]

9 The Fingerprint biometric 9 Finger-scan technology is a widely deployed biometric technology. The fingerprint biometric is highly accurate and versatile. Low-cost and small-size of fingerprint acquisition devices. Figure 4: Smooth skin [1] Figure 5: Friction ridge skin on the fingertips [1]

10 The Fingerprint biometric How it works Fingerprint recognition is feature-based. For optimal feature extraction, the image is enhanced and thinned to one pixel wide. 10 Figure 6: Grayscale fingerprint image [1] Figure 7: Thinned fingerprint image [1] Figure 8: Ridge ending and bifurcation [1]

11 The Fingerprint biometric How it works A minutiae set is an abstract representation of the ridge skeleton. The minutiae set is then matched with a template to compute the match score. 11 Figure 9: Minutiae matching process [1] Figure 10: A genuine pair with maximum matched minutiae [1] Figure 11: An imposter pair with very few matched minutiae [1]

12 The Fingerprint biometric Need for image enhancement Lack of robustness against image quality degradation. Several factors determine the quality of a fingerprint image. In an ideal fingerprint image, ridges and valleys alternate and flow in a locally constant direction. Poor quality images result in spurious and missed features. 12

13 The Fingerprint biometric Need for image enhancement 13 Figure 12: Examples of low quality fingerprint images: (a) dry finger, (b) wet finger, and (c) finger with many creases [1].

14 Fingerprint image enhancement 14 The image enhancement is carried out in either spatial or frequency domain. Spatial domain. g(x,y)=t[f(x,y)] Frequency domain. g(x,y)= τ^(-1) [H(u,v)F(u,v)] Figure 13: Fingerprint image enhancement process [14]

15 Tuned Gabor filtering 15 The state of the art fingerprint enhancement technique is employed by L. Hong et al. [16]. Principle: Convolution of the image with Gabor filter. The filter is tuned to the local ridge orientation and ridge frequency.

16 Tuned Gabor filtering 16 Figure 14: Flowchart of the L. Hong et al. fingerprint enhancement algorithm [13]

17 Tuned Gabor filtering - Normalization Normalization is used to standardize the intensity values in an image. Normalization does not change the ridge structures in a fingerprint. 17 Figure 15: The result of normalization. (a) Input image. (b) Normalized image [13]

18 Tuned Gabor filtering - Orientation image estimation 18 The orientation image is an intrinsic property of the fingerprint images. Figure 16: Orientation estimation at pixel (x,y) [13] Divide the image into blocks of size w x w. Compute gradients Vx(i,j) and Vy(i,j) at each pixel (i,j). Estimate the local orientation of each block centered at pixel (i,j)

19 Tuned Gabor filtering Frequency image estimation 19 Local ridge frequency is another intrinsic property of a fingerprint image. The gray levels along ridges and valleys are modeled as a sinusoidal wave along the local ridge orientation. The image is divided into blocks of size wxw. The x-signature X[0],X[1],,X[l-1] of the ridges and valleys within the window is computed. The frequency of ridges and valleys can be estimated from the x-signature.

20 Tuned Gabor filtering Frequency image estimation 20 w 1 X k = 1 w d=0 G u, v, k = 0,1,., l 1 u = i + ቇ ቆd w 2 coso i, j + ൭k l ൱ 2 sino i, j v = j + ቇ ቆd w 2 sino i, j + ቆ l ቇ 2 k coso i, j Figure 17: Oriented window method to estimate the ridge frequency [13]

21 Tuned Gabor filtering Filtering The estimated frequency and orientation in a fingerprint image provide useful information which helps in removing undesired noise. A bandpass filter that is tuned to the estimation can efficiently remove noise and preserve the true ridge and valley structures. Gabor filters have both frequencyselective and orientation-selective properties. 21

22 Tuned Gabor filtering Filtering The general form of a Gabor filter is: h x, y: φ, f = exp 1 2 x 2 2 δ2 + y x δ2 y cos 2πfx The enhanced image E is obtained by: E i, j = σ u= wg 2 wg σ 2 v= wg 2 h u, v: O i, j, F i, j G i u, j v wg 2 22 Tuned Gabor filtering Figure 18: Original scanned fingerprint [13] Figure 19: Enhanced fingerprint image [13]

23 Related work done 23 Ridge structures that are affected by unusual input contexts can be very complicated. Not all unrecoverable regions can be recovered, as it is difficult to accurately estimate filter parameters in bad sections. A two-stage scheme to enhance the lowquality fingerprint image in both the spatial domain and the frequency domain was proposed by J.Yang et al [9].

24 Two-stage enhancement 24 Figure 20: Flowchart of the two-stage enhancement algorithm [9]

25 Two-stage enhancement Spatial Ridge Compensation 25 The first stage performs ridge compensation along the ridges in the spatial domain. This stage increases the ridge contrast. This stage consists of three steps: Local normalization Local orientation estimation Local ridge-compensation filtering

26 Two-stage enhancement Spatial Ridge Compensation The Local normalization and orientation estimation is similar to the Tuned Gabor filter algorithm [16]. With the orientation of each sub-image estimated, a oriented rectangular window h x w is created. The ridge compensated image is: F i, j = (σ (w 1)/2 m= (w 1)/2 (h 1)/2 σ n= (h 1)/2 (((w 1) β+α) h) N(i,j )) 26 i = i + m cos O i, j j = j m sin O i, j + n sin O i, j + n cos O i, j Figure 21: Window along the local ridge orientation [9]

27 Two-stage enhancement Spatial Ridge Compensation 27 Figure 22: Original fingerprint image [9]. Figure 23: normalized image[9]. Figure 24: First-stage enhanced image[9]

28 Two-stage enhancement Frequency Bandpass Filter The result of the first spatial filter increases the ridge contrast. 28 The frequency bandpass filters used are orientation and frequency selective. The Filter used is the Gabor Filter as in the state of the art approach [16]. Figure 25: Original fingerprint image [9]. Figure 26: First stage enhanced image [9]. Figure 27: Final enhanced image [9].

29 Image pyramids 29 The task Image pyramids or multiresolution processing is to decompose images into multiple information scales. The Laplacian Pyramid Decomposition and Reconstruction (LPD and LPR) are used. The Laplacian pyramid is relevant in this study as all the relevant information is concentrated within a few frequency bands.

30 Image pyramids 30 Image pyramid coding is to low-pass filter the original image go to obtain image g1, which is a reduced version in a way that both resolution and sample density are decreased. In a similar way form g2 as a reduced version of g1, and so on. The sequence g0,g1,,gn is called the Gaussian pyramid. The decomposition method in image pyramids is known as reduce function and reconstruction is known as expand.

31 Image pyramids 31 Take an image g0 with C x R dimensions and an equivalent weighing function h, then the reduced image gl = REDUCE(gl-1) C 2 g l i, j = R 2 m= C n= R 2 2 h l m, n g l 1 2i + m, 2j + n The reconstruction gl,n = EXPAND(g(l,n-1)) C 2 g l,n i, j = R 2 m= C n= R 2 2 h l m, n g l,n 1 i m 2, j n 2 Here, l is the number of pyramid levels 0 < l < n, and (i,j) are pixel values where 0 i < C and 0 j < R [32].

32 Image pyramids 32 Figure 28: First six levels of the Gaussian pyramid for the Lena image [32].

33 Image pyramids Laplacian Image Pyramid Laplacian pyramid representation is the error images which remain when consequent layers of Gaussian pyramid images are subtracted [32]. The LPD images L0,L1.. Ln are obtained by: 33 L l = g l EXPAND g l+1 The reconstruction (LPR) is obtained by: g l = L l + EXPAND g l+1

34 Image pyramids Laplacian Image Pyramid 34 Figure 29: First four levels of the Laplacian pyramid for the Lena image [32].

35 Proposed approach Laplacian pyramid applied to Two-Stage enhancement Enhance the novel two-stage enhancement[9] by adding a LPD stage prior and LPR after. The vital ridge features are present in the 1 st and 2 nd level of the Laplacian pyramid. The original image is decomposed into 2 levels of Laplacian pyramids and each will undergo the Normalization and Gabor filtering. 35

36 Proposed approach Laplacian pyramid decomposition 36 g0 L0 g1 L1 Figure 29: Generation of the Laplacian pyramid decomposed images. g2

37 Proposed approach Laplacian pyramid applied to Two-Stage enhancement 37 Original Image Laplacian Pyramid Decomposition L0 Two-Stage Enhancement Laplacian Pyramid Reconstruction Enhanc ed image L1 Two-Stage Enhancement Figure 30: Proposed Multi-stage enhancement scheme

38 Results 38 Figure 31: Input FVC2004DB1 102 image [18] Figure 32: Level 1 Laplacian Pyramid Image Figure 33: Level 2 Laplacian Pyramid Image Figure 34: Level 1 Laplacian Pyramid enhanced Image Figure 35: Level 2 Laplacian Pyramid enhanced Image Figure 36: Final enhanced image

39 Results 39 Figure 37: Input FVC2004 DB1 105 image [18] Figure 38: State of the art enhanced image Figure 39:Two stage enhanced image Figure 40: LPD based multi-stage enhancement image

40 Results 40 Figure 42: State of the art enhanced image with distortion highlighted Figure 41: Input FVC2004DB1 102 image [18] Figure 43: LPD based multi-stage enhanced less distortion highlighted

41 Results 41 Figure 44: Input image of very bad quality FVC2004DB1 101 [18] Figure 45: State of the art enhanced image Figure 46: LPD Multistage scheme enhanced image

42 Results - Thinning 42 Input image Input image Input image Thinned Image Thinned Image Thinned Image Figure 47: Input FVC2004DB2 101 and thinned image [18] Figure 48: State of the art enhanced and thinned image Figure 49: LPD Multistage scheme enhanced and thinned image

43 Results - Minutiae 43 Minutiae Minutiae Minutiae Figure 50: Original image minutiae[18] Figure 51: State of the art enhanced image minutiae Figure 52: LPD Multistage scheme enhanced image minutiae

44 Results Minutiae ratio and time 44

45 Results Minutiae ratio 45

46 Results - Minutiae 46

47 Conclusion and Future Work The fingerprint biometric is a feature based biometric system. The feature extraction heavily relies on optimal image enhancement. A novel multi-stage enhancement scheme has been proposed which by using LPD and LPR, gives better fidelity to the ridge-valley structure of the fingerprint images. Due to the multi-tier and multi-stage approach the performance time has increased which can be vital for a real time biometric system. Future work: Optimize the Bandpass filter for the frequency domain. Use an efficient multi-threaded programming language. 47

48 Test Conditions 48 The code was developed and executed using the Mathworks MATLAB R2013a tool which uses a proprietary programming language and has support for image processing using its image processing toolbox. The fingerprint recognition v2.2 software is used for true minutiae count [36] and the Fingerprint minutiae extraction matlab code [37] to compute the total minutiae count and, ratios TMR and FMR. All the work was done on a system with following configuration: Operating System: Windows 10 Home Edition Processor: Intel(R) Core(TM) i3-3120m 2.50GHz 2.50GHz RAM: 8.00 GB System type: 64-bit Operating System, x64-based processor

49 Acronyms 49 AFIS - Automated Fingerprint Identification Systems DNA - Deoxyribonucleic acid FMR - False minutiae ratio FVC - Fingerprint verification completion LPD - Laplacian pyramid decomposition LPR - Laplacian pyramid reconstruction PIN - Personal Identification Number TMR - True minutiae ratio

50 References 50 [1] A. K. Jain, A. A. Ros, and K. NandaKumar, Introduction to Biometrics, Springer, [2] S. C. Lee, An Introduction to Identity Management, SANS Institute, March 2003 [3] Article by V.Maty aˇs and Z.R ıha on Biometric Authentication Security and Usability : [4] A. K. Jain, A. Ross, and S. Pankanti, Biometrics: A tool for information security, IEEE Trans. Info. Forensics Security, vol. 1, no. 2, pp , June [5] Article on the Overview of Biometrics: the- [6] Article on the future of Biometrics: [7] Article on types of Biometrics:

51 References 51 [8] D. Petrovska-Delacrétaz, G. Chollet, and B. Dorizzi, Guide to Biometric Reference Systems and Performance Evaluation, Springer- Verlag, [9] J.Yang, N. Xiong, and A. V. Vasilakos Two-stage enhancement scheme for lowquality fingerprint images by learning from the images IEEE transactions on Human-Machine Systems, vol. 43, no. 2, pp , Mar [10] A. K. Jain, Fundamentals of Digital Image Processing, Englewood Cliffs, NJ: Prentice Hall, [11] R.C. Gonzalez and R.E. Woods, Digital Image Processing, 3rd edition, Prentice Hall, [12] A. K. Jain, A. Ross, and S. Prabhakar, An Introduction to Biometric Recognition, IEEE Trans. Circuits and Systems for Video Technology, vol. 14, no. 1, pp. 4 20, Jan [13] L. Hong, Y. Wan, and A.K. Jain, Fingerprint image enhancement: Algorithm and performance evaluation. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20, no. 8, pp , Aug

52 References 52 [14] D.K. Misra, S.P. Tripathi and D. Misra, A Review Report on Fingerprint Image Enhancement Techniques. International Journal of Emerging Trends & Technology in Computer Science, vol. 2, no. 2, pp , April [15] S. Greenberg, M. Aladjem, D. Kogan, and I. Dimitrov, Fingerprint image enhancement using filtering techniques, ICPR, Vol. 3, pp , Sep [16] L. Hong, A.K. Jain, S. Pankanti, and R. Bolle, Fingerprint Enhancement, Proc. First IEEE WACV, pp , Sarasota, Fla.,1996. [17] S. Chikkerur, A. N. Cartwright, and V. Govindaraju, Fingerprint enhancement using STFT analysis, IEEE Transactions on Pattern Recognition, vol. 40, no. 1, pp , Jan [18] Fingerprint Verification Competition database: [19] Article on Gabor filter: filter.html

53 References 53 [20] K.R.Rao, D.N.Kim and J.J.Hwang, Fast Fourier transform: Algorithms and applications, Heidelberg, Germany: Springer [21] MS thesis paper by S. Chakraborty on Fingerprint enhancement by directional filtering : 6.pdf [22] R.C. Gonzalez, R.E. Woods and S.L. Eddins, Digital Image Processing Using MATLAB, 2nd edition, McGraw Hill Education, [23] Z. Shi and V. Govindaraju, A chaincode based scheme for fingerprint feature extraction, IEEE Transactions on Pattern Recognition, vol. 27, no. 5, pp , Apr [24] K.G.Darpanis, Gabor Filters, York University, April [25] O. Marques, Practical Image and Video Processing Using MATLAB, 1st Edition, Wiley-IEEE Press, [26] Matlab:

54 References 54 [27] Multimedia processing course website: [28] Visual studio: [29] D. Kumar and Y. Ryu, A Brief Introduction of Biometrics and Fingerprint Payment Technology, International Journal of Advanced Science and Technology, vol. 4, pp , March [30] P.K. Shimna and B. Neethu, Fingerprint Image Enhancement Using STFT Analysis, IOSR Journal of Electronics and Communication Engineering, vol. 10, Issue 2, pp , Apr [31] D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, Handbook of fingerprint recognition. Springer, [32] P.J. Burt and E.H. Adelson, The Laplacian Pyramid as a Compact Image Code, IEEE Transactions on communications, vol. 31, no. 4, pp , Apr [33] H. F. D. Kunz, K. Eck, and T. Aach, "A Nonlinear Multi-Resolution Gradient- Adaptive Filter for Medical Images", In SPIE Medical Imaging, volume 5032, pages , May 2003.

55 References 55 [34] A. Jepson, Notes on Image Pyramids to attenuate noise : [35] H. Fronthaler, K. Kollreider, and J. Bigun, Local features for enhancement and minutiae extraction in fingerprints, IEEE Trans. Image Process., vol. 17, no. 3, pp , Mar [36] Fingerprint Recognition v2.2 software : Fingerprint-Verification. [37] PowerPoint presentation background image:

56