FINGERPRINT RECOGNITION AKSHAY KANCHAR25 DHRUV PANCHAL44 RICHA SAXENA57

FINGERPRINT RECOGNITION AKSHAY KANCHAR(25) DHRUV PANCHAL(44) RICHA SAXENA(57) PROJECT GUIDE : - K. T. TALELE BHARTIYA VIDYA BHAVAN'S SARDAR PATEL INSTITUTE OF TECHNOLOGY MUNSHI NAGAR, ANDHERI (W), MUMBAI - 400 058. 2011 -12

Overview Of The Presentation �Introduction �Literature Survey Summary �Problem Definition �Proposed System �Experimental Results �Conclusion 2

Fingerprint Features Core Point Ridge Ending Bifurcation 3

Overview Of The Presentation �Introduction �Literature Survey Summary �Problem Definition �Proposed System �Experimental Results �Conclusion 4

Literature Survey (1/3) [1] “Fingerprint Recognition using Minutiae Score Matching (FRMSM)”, Ravi. J, K. B. Raja, Venugopal. K. R, 2009. �This paper mainly focuses on minutiae extraction. �Overlap error correction block is absent. �Minutiae matching method used is very complex. �Efficiency: � Probability of rejecting approved user=0% � Probability of accepting false user=2. 6% 5

Literature Survey (2/3) [2] “Robust Fingerprint Recognition System using Orientation and Texture features”, Zin Mar Win, Myint Sein, 2012. �Entire fingerprint isn’t used. Only the region close to the core point is cropped and used for further computations �Minutiae is not used as a feature of the fingerprint. �They filter the fingerprint image in 8 directions and generate a fingercode for each direction which is used for matching. �Efficiency: Probability of rejecting approved user=1. 5% � Probability of accepting false user=1. 7% � 6

Literature Survey (3/3) [3] “Fingerprint recognition by Euclidian Distance”, Chomtip Pornpanomchai, Apiradee Phaisitkulw, 2010 �This paper solely concentrates on minutiae extraction and matching. Other blocks of preprocessing are not mentioned. � 52 possible combinations of 3 x 3 pixel masks are used to extract minutiae points. This isn’t necessary as it can be done by using Crossing Number Technique. �Distance between minutiae points and the core point is used as the feature for matching process. �Efficiency: Probability of rejecting approved user=5% � Probability accepting false user=0% � 7

Overview Of The Presentation �Introduction �Literature Survey Summary �Problem Definition �Proposed System �Experimental Results �Conclusion 8

Literature Survey Summary �Pros: �Papers [2] and [3] use only the region around the core point which reduces the computations and increases the speed of the system �Paper [3] uses distance between the core point and minutiae points as the parameter for matching. This eliminates the need for rotating the image along the orientation of the template image �Paper [1] has the least probability of rejecting an approved user �Paper [3] has the least probability of accepting a false user 9

Literature Survey Summary �Cons: �The orientation estimation algorithm used were complicated. �Filtering process for image enhancement are complex. �Overlap error correction is not provided by the existing algorithms. This leads to the formation of false minutiae points. 10

Overview Of The Presentation �Introduction �Literature Survey Summary �Problem Definition �Proposed System �Experimental Results �Conclusion 11

Problem Definition �Fingerprint Recognition Using Minutiae Extraction and Euclidean Distance to, �Accept query image with proper dimensions and white background �Convert the image to grayscale �Segment the image into foreground and background �Normalize the image so that the pixels lie within a particular range �Determine the orientation of the foreground blocks �Filter the foreground blocks in their respective orientations 12

�Binarize the image �Remove overlap error introduced by filtering stage �Obtain thinned image of the overlap corrected image �Extract minutiae points by Crossing Number technique �Extract core point using a 9 x 9 pixel mask �Calculate Euclidean Distance of 32 points closest to the Core point �Determine the matching score 13

Overview Of The Presentation �Introduction �Literature Survey Summary �Problem Definition �Proposed System �Experimental Results �Conclusion 14

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 15

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 16

Image Acquisition �File Formats : . tif, . jpeg, . bmp, . png �Recommended sizes : 360 x 480, 480 x 640 �Images resolution preferred in multiples of 30 �Image DPI(Dots per inch) Range : 400<DPI<600 �Image Colour types : Colour formats (8, 16, 24, or 32 bits), Grayscale formats(8 or 16 bits) 17

Image Acquisition �Image background should be white. �Fingerprint should not be at the edge of the image. There should be proper spacing along all 4 sides of the image. �Examples: - 18

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 19

Step: - 1 Segmentation Input Image Segmentation using Variance Threshold method Segmented Image 20

�Method based on variance threshold is used. �If (variance<threshold) …………………. Threshold=40 �Block is a part of Background �If (variance>=threshold) �Block is a part of Foreground �To calculate variance, 21

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 22

Step: - 2 Normalization Input Image Normalization using Variance , Mean Threshold method Normalized Image 23

�Normalization is done so that a common threshold value for binarization can be applied to all the fingerprint images. �The normalized image is defined as, I(i, j)>M Where M 0=Desired Mean of pixel intensity=100 Vo=Desired Variance of a block=1500 24

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 25

Step: - 3 Orientation Estimation Normalized Image Orientation Estimation using a 9 x 9 pixel mask Orientation Map 26

�First we decide the ridge direction of each pixel 27

�Compute the average grey value in 8 directions. (sum[i]) �Form groups of 2 having directions i and i+4. �Compute the difference Gd[i]=|sum[i]sum[i+4]|……………. i=0, 1, 2, 3 �One of the directions of the pair having max difference is the ridge direction. (imax) �D=imax, if(|grey-G[imax]|<|grey-G[imax+4]|) =imax+4 otherwise 28

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 29

Step: -4 Gabor Filtering Result Normalized Image Filtering using Gabor Filtered Image 30

�Gabor filters are employed because they have frequency selective and orientation selective properties. �These allows filter to give maximal response to ridges at a particular orientation �Preserves the ridge structure and reduces noise 31

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 32

Step: - 5 Binarization Filtered Image Binarization using Threshold method Binarized Image 33

�Only two levels: Black pixels that represent ridges and white pixels that represent valleys �If (pixel value<threshold)……………… Threshold=150 �Pixel = 0 �If (pixel value>=threshold) �Pixel=255 �Binarization improves the contrast between the ridges and valleys. 34

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 35

Step: - 6 Overlap Error Elimination Binarized Image Overlap Error Elimination Overlap Error. Free Image 36

Thinning & Minutiae Extraction Overlap Error Correction Thinning & Minutiae Extraction 37

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 38

Step: - 7 Thinning result Overlap Error. Free Image Thinning using Guo. Hall algorithm Thinned Image 39

�Thinning is the process of reducing the thickness of an object in a digital image to the minimum size necessary for machine recognition of that object. �The output of the thinning process is known as the “skeleton” hence also referred to as skeletonization. 40

Thinning Conditions �Where: C(P 1) = !P 2 & (P 3 | P 4) + !P 4 & (P 5 | P 6) + !P 6 & (P 7 | P 8) + !P 8 & (P 1 | P 2) N 1(P 1) = (P 9 | P 2) + (P 3 | P 4) + (P 5 | P 6) + (P 7 | P 8) N 2(P 1) = (P 2 | P 3) + (P 4 | P 5) + (P 6 | P 7) + (P 8 | P 9) N(P 1) = MIN[N 1(P 1), N 2(P 1)] 41

Thinning Algorithm • We have implemented Guo-Hall algorithm for thinning. • While points are deleted do • For all pixels p(i, j) do if (a) C(P 1) = 1 (b) 2 ≤ N(P 1) ≤ 3 (c) Apply one of the following: 1. (P 2 | P 3 | !P 5) & P 4 = 0 in odd iterations 2. (P 6 | P 7 | !P 9) & P 8 = 0 in even iterations then Delete pixel p(i, j) end if end for end while 42

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 43

Step: - 1 Minutiae Extraction Thinned Image Minutiae Extraction using Crossing Number Technique Minutiae Map 44

45

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 46

Step: - 2 Core Point Detection Orientation Map Core Point Detection Using 9 x 9 Pixel Mask Thinned Image 47

�We use a 9 x 9 mask to determine the core point. �A pixel is said to be a core point if the Loop Field Strength is minimum. �If we interchange +1 and -1 then the core point is the pixel having maximum Loop Field Strength 48

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 49

Step: - 3 Euclidean Distance Detection Thinned Image Applying Euclidean Distance b/w Core Point And Minutiae Points Euclidean Distance 50

�Divide the area around the core point into 8 sectors. �Calculate the Euclidean distance of minutiae points from the core point. �Select 2 points from each sectors having the least Euclidean Distance. �These 16 distances are used for matching between query image and template image. 51

Proposed System Pre-processing Image Acquisition • Acquire Image • Convert to Grayscale • Segmentation • Normalization • Orientation Estimation • Filtering • Binarization • Overlap Error Elimination • Thinning Final Processing • Minutiae Extraction • Core Point Extraction • Euclidean Distance • Matching Score 52

Step: - 4 Matching Score �We arrange the 16 Euclidean Distances in ascending order. �Then we compare these distances one is to one with the database. �Matching score is calculated based on the number of points matched out of the 16 points. 53

Overview Of The Presentation �Introduction �Literature Survey Summary �Literature Survey Conclusion �Proposed System �Experimental Results �Conclusion 54

Experimental Result 1 Query Image Template Image 55

Experimental Result 2 Query Image Template Image 56

Overview Of The Presentation �Introduction �Literature Survey Summary �Literature Survey Conclusion �Proposed System �Experimental Results �Conclusion 57

Conclusion �Here we presented Fingerprint matching using minutiae extraction and comparing euclidean distances between the core point and minutiae points. �In pre-processing stage the steps of segmentation, normalisation, orientation estimation, filtering, binarization are performed to enhance the image. �In processing stage the steps of thinning, minutiae extraction, core point extraction are performed. �In post-processing stage the euclidean distance between the core point and minutiae points are found out. �These euclidean distances are used for matching query image with the database therefore eliminating the need of aligning the test image with the database image. 58

References �Ravi. J, K. B. Raja, and Venugopal. K. R, `` Fingerprint recognition using minutia score matching '', International Journal of Engineering Science and Technology Vol. 1(2), 2009, 35 -42 � Zin Mar Win and Myint Sein, ``Robust Fingerprint Recognition System using Orientation and Texture features '', 2 nd International Conference on Computer Science and Information Technology (ICCSIT'2012) Singapore April 28 -29, 2012 �Mar Min and Yadana Thein, `` Intelligent Fingerprint Recognition System by Using Geometry Approach '', Proceedings of IEEE Conference, 2009 59