(IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 A Multistage Detection and Elimination of Spurious Singular Points in Degraded Fingerprints Zia Saquib, Santosh Kumar Soni, Sweta Suhasaria Dimple Parekh, Rekha Vig Center for Development of Advanced Computing NMIMS University, Mumbai, Maharashtra 400049, India Mumbai, Maharashtra 400056, India firstname.lastname@example.org email@example.com firstname.lastname@example.org email@example.com firstname.lastname@example.org Abstract—Singular point (SP) detection is one of the most gives the genuine set of SPs. These methods, the proposed crucial phases in fingerprint authentication systems and is scheme and its comparison with one of the state-of-the-art used for fingerprint classification, alignment and techniques are explained in detail in section II. Experimental matching. This paper presents a multistage approach for results are discussed in sections III, followed by conclusion in detection and elimination of spurious singular points section IV. especially in degraded fingerprints. The approach II. THE PROPOSED SCHEME AND ITS KEY COMPONENTS comprises three stages. In the first stage, two different methods, viz., quadrant change and orientation reliability A. Quadrant Change: Method-A measure, are independently employed on the same image As per K. Kryszczuk and A. Drygajlo (2006), a singular point to generate two sets of candidate singular points. The is the location where the general ridge orientation becomes second stage performs the multiscale analysis on a set of discontinuous. Informally, this can be stated as the area where candidate SPs located by reliability method, which ridges oriented rightwards change to leftwards and those that improves the approximation by reducing the list of SPs. In were oriented upwards turn downwards, and opposite. This the third stage, the spurious singular points are detected information can be extracted from the quadrant change of the and thereby eliminated by taking the intersection of the averaged square gradients. The orthogonal gradient two sets of SPs. This model is tested on a proprietary components in the x and y directions are considered separately. (Lumidigm Venus V100 OEM Module sensor) fingerprint In general, each pair of corresponding gradient components database at 500 ppi resolution. The experimental results manifests the gradient quadrant change by the change of sign. show that the approach effectively eliminates the spurious The sign maps PMx and PMy are computed using the Eq. (1): SPs from the noisy and highly translated/rotated fingerprint images. The proposed scheme is also compared with one of the state-of-the-art techniques, the experimental results prove its superiority over the later. (1) Keywords- Spurious Singular Points, Multiscale Analysis, Orientation Consistency, Quadrant Change, Reliability, Minimum Inertia, Maximum Inertia. We need to locate points in whose respective local ridge I. INTRODUCTION gradients change sign in both x and y directions. These points are obtained by computing the intersection of the two sets of The performance of fingerprint authentication system has such points for which the sign of the y-directional and x- come a long way but it is still influenced by many factors, like: directional (respectively) gradient component changes, as inaccurate detection of singular points (core and delta). Poor- shown in Eq. (2): quality and noisy fingerprint images mostly result in false or missing singular points (SPs), which generally results in degradation of the overall performance of the authentication (2) systems. This paper presents a three-stage approach, which primarily focuses on the detection and elimination of spurious The operator edge in Eq. (2) denotes any edge detector that SPs for all types of fingerprint images, especially noisy images. works on binary images, and [xsp, ysp] are the points where two This paper puts forward an effective way to locate a unique quadrants change boundaries intersect, as shown in Figure 1. reference point consistently and accurately using tri-method [xsp, ysp] are considered as SPs, as shown in Figure 2. This fusion scheme. Method-A works on the quadrant change method works well with good quality gray-level images, but information, whereas, Method-B uses pixel-wise reliability the moment image quality degrades, it starts resulting in measure of the orientation field followed by multiscale analysis spurious SPs and eventually becomes ineffective, as shown in to compute candidate SPs. Intersection of methods A and B Figure 2. (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 Figure 1. Horizontal and Vertical maps. Figure 3. Reliability Image Figure 4. Genuine and Spurious SPs based on Reliability Measure. Figure 2. Genuine and Spurious SPs based on Quadrant Change Information. C. MultiScale Analysis B. Orientation Reliability Measure: Method B As per T. Van and H. Lee (2009), a multiscale analysis As per Z. Saquib and S. K. Soni (2011), M. Khalil, D. (see Figure 5) of orientation consistency is used to search the Muhammad (2010), the raw fingerprint image is first filtered local minimum orientation consistency from large scale to fine using Gabor filter. Then, 'reliability' of ridge orientation map scale. The orientation consistency-based technique can be is calculated, followed by the calculation of the area of summarized as follows: moment of inertia about the orientation axis (the min. inertia) 1) Compute the orientation consistency Cons(s) of each block and an axis perpendicular (the max. inertia), as given in Eq. based on the outside 8s surrounding blocks of its (2s+1) x (3) and (4): (2s+1) neighborhood. 2) Find the minimum orientation consistency denoted as min_inertia(x, y) = (((Gyy + Gxx) - (Gxx - Gyy) * φ'x) - (Gxy * φ'y))/2 (3) Consmin (s). Compute candidate threshold as, max_inerita(x, y) = Gyy + Gxx – min_inertia(x, y) (4) (6) where, φ'x and φ'y are cosine and sine of doubled angles (ridge orientations). The reliability measure is given by Eq. (5): 3) Select the blocks if their Cons(s) < T. Reliability Measure = 1.0 – min_inertia/max_inertia (5) 4) Compute dx(s) and dy(s), and select the blocks with both dx(s) and dy(s) larger than 0 as the candidate blocks in All such pixels with reliability measure below an empirically the next finer scale: determined threshold (here, it is 0.035) are considered as the candidate SPs. The pixels with deep blue shades are the possible SPs, as shown in Figure 3, and the corresponding SPs are shown in Figure 4, which is inclusive of both genuine and spurious. (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 cases are also presented in Figure 9, where the raw images (7) chosen are relatively of much poorer quality than the images in Figure 8. (8) III. EXPERIMENTAL RESULTS Proprietary (Lumidigm Venus V100 OEM Module sensor) 5) If no candidate blocks for the reference point are located, dataset has been chosen as test data to evaluate the impact of let T = T + 0.01, go to step 3). the proposed multistage scheme for detection and elimination 6) Repeat steps 1), 2), 3), 4), and 5) in the selected candidate of spurious SPs. The scheme is implemented in MATLAB. The blocks with s = s-1 until s = 1. experimental results show that this approach satisfactorily 7) Locate the block with minimum orientation consistency improves the accuracy of detection of correct singular points in Cons(1) from the selected finest scale blocks as the noisy and highly transformed (translated/rotated) fingerprint unique reference point. images. Only select cases (highly degraded/translated/rotated) have been chosen to measure the effectiveness of the approach. Few of them are presented in Figure 7 and 8. Some improved cases are also displayed, as shown in Figure 9, where severely distorted/poorly overlapped fingerprint images are chosen, which present real challenges in the fields. IV. CONCLUSION Genuine SPs are very crucial towards attaining high accuracy and performance of the authentication systems. Thus, spurious SPs need to be completely removed. In this paper, a multistage scheme is proposed for detection and elimination of spurious singular points, especially in highly degraded, translated and rotated fingerprint images. Experimental results clearly show that the three methods in combination effectively remove (or minimize) the spurious singular points. The scheme is tested against some select difficult cases. Also, this method (fourth column in Figure 8), upon comparison with the Figure 5. The multiscale analysis of orientation consistency. approach presented by Z. Saquib, S. K. Soni (2011) (second column in Figure 8), is found better. We have performed multiscale analysis over the set of SPs given by reliability measure stage for better approximation of ACKNOWLEDGMENT the genuine SPs, as explained in sub-section D. Multiscale We wish to extend our sincere thanks to the Department of analysis helps in reducing the list of SPs further by isolating Information Technology (DIT), Ministry of Communications and removing the false SPs. and Information Technology, Govt. of India, for assigning us a D. Proposed Approach: A Multistage Detection and biometric project: “BharatiyaAFIS”. This work is carried out as a part of the same project. Elimination of Spurious SPs The proposed approach, as shown in Figure 6, comprises the REFERENCES state-of-the-art methods (with some modifications/tuning)  T. Van and H. Lee,“An efficient algorithm for fingerprint reference- presented in sub-sections A, B and C. Firstly, the two sets of point detection”, IEEE 2009. candidate SPs are generated using the methods: i) quadrant  K. Kryszczuk and A. Drygajlo, “Singular point detection in fingerprints change information and ii) reliability measure of the orientation using quadrant change information”, The 18th International Conference field. In order to have better approximation, multiscale analysis on Pattern Recognition (ICPR'06), 2006. is performed over the candidate SPs from reliability measure,  D. Maltoni, D. Maio, A. Jain, and S. Prabhakar, Handbook of which reduces (or minimizes) the list by identifying, and Fingerprint Recognition. New York: Springer, 2003. thereby ignoring most of such pixels which are not likely to be  L. Hong, Y. Wan, and A. Jain, “Fingerprint image enhancement: algorithm and performance evaluation”, IEEE Transactions On Pattern the SPs. Finally, the genuine SPs are confirmed by taking the Analysis And Machine Intelligence, Vol. 20, No. 8, 1998. intersection of the two sets of SPs from the above two methods,  M. Khalil, D. Muhammad, M. Khan, Mohammed, “Singular points which then filters out the false SPs, if any, leaving behind detection using fingerprint orientation field reliability”, International genuine SPs. These stages are shown together in Figure 6. The Journal of Physical Sciences Vol. 5(4), pp. 352-357, 2010. experimental results are shown in Figure 7 and 8. In Figure 8,  Z. Saquib, S. Soni, S. Suhasaria, D. Parekh, R. Vig, “A fault-tolerant first column depicts the raw images, second column shows the approach for detection of singular points in noisy fingerprint images”, results using Quality Change and Reliability methods, third International Journal of Computer Security Issues, Volume 8, 2011. column displays SPs by Quadrant Change Information (blue),  http://en.wikipedia.org/wiki/Euclidean_distance Reliability Measure (red), Multiscale Analysis (green) and the  Kovesi PD (2008). MATLAB and Octave Functions for Computer fourth column presents results from the proposed scheme Vision and Image Processing, in School of Computer Science and Software Engineering, The University of Western Australia. Available (genuine SPs are depicted by orange color). Few improved from http://www.csse.uwa.edu.au/~pk/research/matlabfns/. (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 Figure 6. Proposed Scheme. Genuine SP Genuine SP Spurious SP Figure 7. SPs before Intersection (left), SPs after Intersection (right). (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 Quadrant Change, Reliability & Quadrant Change, Reliability Quadrant Change & Fingerprint Image Multiscale methods & Multiscale methods Reliability methods (before Intersection) (after Intersection) 001_5_10.bmp 001_5_68.bmp 003_5_73.bmp 006_5_16.bmp (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 006_5_34.bmp 006_5_55.bmp 006_5_60.bmp 006_5_75.bmp (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 007_5_2.bmp 007_5_23.bmp 007_5_67.bmp 001_5_26.bmp Figure 8. Experimental Results from Lumidigm Dataset: (first column) Raw Images, (second column) Results using Quality Change and Reliability methods, (third column) Blue SPs by Quadrant Change Information, Red SPs by Reliability Measure, Green SPs by Multiscale Analysis and (fourth column) Proposed Scheme – Genuine SPs are depicted by Orange SPs. (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No.5, May 2011 Quadrant Change, Reliability Quadrant Change & Reliability Sr.No. Fingerprint Image & Multiscale methods methods (proposed approach) 1. 006_5_65.bmp (There is no SP present in the Raw Image) 2. 006_5_66.bmp (Only Delta should have been marked) 3. 007_5_25.bmp (Only single Core is present) 4. 001_5_15.bmp (Only single Core is present) Figure 9. Experimental Results from Lumidigm Dataset: Third column represent improved cases, inclusive of both genuine and spurious SPs (please zoom to view them properly).