Enhancement and Minutiae Extraction Of Touchless Fingerprint Image Using Gabor And Pyramidal Method

Document Sample
Enhancement and Minutiae Extraction Of Touchless Fingerprint Image Using Gabor And Pyramidal Method Powered By Docstoc
					                                                         (IJCSIS) International Journal of Computer Science and Information Security,
                                                         Vol. 9, No. 3, March 2011

         Enhancement and Minutiae Extraction of
         Touch less Fingerprint Image Using Gabor
                  and Pyramidal Method

                 A.John Christopher
Associate Professor, Department of Computer Science,                                         Dr.T.Jebarajan,
            S.T. Hindu College, Nagercoil                                                       Principal,
                                                                           V.V. College of Engineering., Tisayanvilai
Abstract - Touch based sensing techniques generate lot of errors
in fingerprint minutiae extraction. The solution for this problem          deformation, slippage, smearing or sensor noise. Some of the
is touchless fingerprint technology. They do not receive any               touch based are shown in fig.1. A new generation of
contact between the sensor & finger. Although they reduce the              touchless live scan devices that generate three various
problems of touch based finger prints, other difficulties explore
such as a view difference problem and a limited usable area due
                                                                           representation of fingerprint is appearing in the market. This
to perspective distortion. To solve this problem, proposed                 new sensing technology addresses many of the problems
method for touchless fingerprint image enhancement and                     stated above [3]. From wear and tear of surface coating, to
minutiae extraction is introduced. Image enhancement is mostly             overcome these kinds of problems, a touchless fingerprint
required preprocessing system for finger based biometric                   sensing technology has been proposed that does not require
system. Normally the touchless device is having a single camera            any contact between a sensor and a finger. Thus, the fingers
and two planer mirrors which reflecting side views of a finger.            and ridge information cannot be changed or distorted as it
From this we get three images normally frontal, left and right             will be free of skin deformation. Also, it can capture
finger. Experimental result shows that the enhanced images                 fingerprint images consistently because it is not affected by
increase the biometric accuracy.
                                                                           different skin conditions or latent fingerprints.
Index Terms - pyramidal method, Gabor, touchless fingerprint,
thinning, normalization, finger enhancement, adaptive histogram.

                      I ‐ INTRODUCTION 

          A fingerprint is composed of ridges and valleys.
Ridges have various kinds of discontinuity such as ridge
bifurification, ridge endings, short ridges, islands and ridge
cross over’s. Among this discontinuity, ridge bifurification
and ridge ending are commonly used in fingerprint
identification/verification system and are called minutiae
[1].For the processing of fingerprint images, two stages are of
pivotal importance for the success of biometric
reorganization: image enhancement and minutiae extraction.
The traditional fingerprint processing technologies are
applied immediately after sensing. But a better thing is an
optional image enhancement in fingerprint images. In
                                                                                 Fig. 1: Distorted images acquired from a touch-based sensor.
realistic scenarios though the quality of a fingerprint image
may suffer from various impairments, caused by scores, cuts,
                                                                           Recently, several companies and research groups have
moist or dry skin, sensor noise, blur, wrong handling of
                                                                           developed touchless fingerprint sensors and recognition
sensor, weak ridge and valley pattern of the given fingerprint,
                                                                           systems [4]–[6]. TST Group developed a touchless imaging
etc. The task of the fingerprint enhancement is to counteract
                                                                           sensor (BiRD III) which uses a complementary metal–
the aforesaid quality impairments and to reconstruct the
                                                                           organic–semiconductor (CMOS) camera, and red and green
actual fingerprint pattern as trace to it original as possible. [2]
                                                                           light sources to acquire fingerprint images [4]. Song et al. [5]
Fingerprints are traditionally captured based on contact of the
                                                                           proposed a sensing system with a single charged-coupled
finger on paper or a platen. This often results in partial or
                                                                           device (CCD) camera and double ring-type blue illuminators
degraded images due to improper finger placement, skin
                                                                           to capture high contrast images. Also, Mitsubishi Electric
                                                                           Corporation proposed another touchless approach

                                                                      52                                     http://sites.google.com/site/ijcsis/
                                                                                                             ISSN 1947-5500
                                                      (IJCSIS) International Journal of Computer Science and Information Security,
                                                      Vol. 9, No. 3, March 2011

transmitting the light through the finger [6], acquiring               can capture three different views of a fingerprint using only
fingerprint patterns under the surface of skin using light with        one Camera and also avoid the synchronization problem
a wavelength of 660 nm. However, such sensing systems [4]–             existing in multiple camera-based systems. In addition, to
[6] have an inherent problem as they use only a single                 obtain high-quality fingerprint images, we need to consider
capturing device, such as CMOS or CCD cameras. when                    several optical components in order to design the device.
capturing an image using a single camera, the geometrical
resolution of the fingerprint image decreases from the
fingerprint center towards the side area [7]. Therefore, false
features may be obtained in the side area and it reduces the
valid and useful region for authentication. Moreover, if there
is a view difference between images due to finger rolling, it
reduces the common area between fingerprints and degrades
system performance. To solve this problem, 3-D touchless
sensing systems using more than one view have been
explored [8]–[11]. TBS [8] used five cameras placed around
a finger to capture nail-to-nail fingerprint images and                                          Fig. 2: Proposed device.
generated a 3-D fingerprint image using the shape-from-                     (a)    Prototype of the device. (b) Schematic view of the device.
silhouette method. They then unwrapped the 3-D finger
image onto a 2-D image by using parametric and                         The specifications of the optical components are as follows:
nonparametric models to make rolled-equivalent images [9].                 1) Camera and lens: We use a 1/3-in progressive scan
Fatehpuria et al. [10] proposed a 3-D touchless device using                   type CCD with 1024 x 768 active pixels, where the
multiple cameras and structured light illumination (SLI). The                  pixel size is 4.65 x 4.65 m. This camera offers a
structured light patterns are projected onto a finger to obtain                sufficient frame rate of 29 Hz, thus avoiding image
its 3-D shape information and 2-D unfolded images are                          blurring caused by typical finger motion. Also, we
generated by applying “Springs algorithm” and some post                        use simple equations [see (1) and (2)] to design an
processing steps. Also, the Hand Shot ID system was                            adequate lens for our system.
developed to acquire a 3-D shape of a hand with fingers by
stitching images from 36 cameras [11]. Although all these
methods attempted to solve the problems in touch-based                                  M =                               (1) 
sensors and acquire expanded fingerprint images with less                                 p
skin deformation, they did not raise much interest in the
                                                                                        1 1 1
market because of much higher costs compared to                                          = +                               (2)
conventional touch-based sensors. Considering the above                                 f p q
observations, we adopt a new touchless sensing scheme using
a single camera and a set of mirrors. The mirrors work as                         Where f is the lens focal length, p and q are the lens-
virtual cameras, thus enabling the capture of an expanded                         to-object and lens-to-image distances, respectively,
view of a fingerprint at one time without using multiple                          and M is the optical magnification. Normally, the
cameras. The device consists of a single camera, two planar                       required image resolution for touch-based sensors is
mirrors, light-emitting diode (LED)-based illuminators, and a                     500 dpi. Therefore, to ensure a 500-dpi spatial
lens. Two planar mirrors are used to reflect the left and right                   resolution in the fingerprint area and to cover three
side view of a finger. In this paper, we proposed a new                           view fingerprints, the optical magnification
method to enhance the touchless finger print and to extract                       parameter M, the lens to image distance, and field of
the minutiae data.                                                                view (FOV) are determined as 0.1, 170 mm, and 50
                    II – SYSTEM DESIGN                                            x 38 mm, respectively. By doing this, we can
                                                                                  capture three view images with 500-dpi resolution at
          To overcome the view difference problem and the                         one time. Also, the depth of field (DOF) of the lens
limitation of a single view, some touchless fingerprinting                        ranges from -2.6 to +2.6 mm at a given working
systems capture several different views of a finger by using                      distance and it normally covers the half depth of a
multiple cameras. However, using multiple cameras increases                       finger.
the cost and size of a system. Thus, we adopt a new sensing
system which captures three different views (frontal, right,               2)      Illumination: Considering the reflectance of human
and left) at one time by using a single camera and two planar                     skin to various light sources, we used ring-shaped
mirrors. Figs. 2(a) and (b) show the prototype and schematic                      white LED illuminators and a band pass filter which
view of the device. As shown in Fig. 2, two mirrors are                           can transmit green light to enhance the ridge-to-
placed next to the finger and reflect the right and left side                     valley contrast. Also, the illuminators are placed
views of the finger. Then, the frontal view and two mirror-                       perpendicular to the finger to remove the shadowing
reflected views are captured by a single camera                                   effect. Diffusers are used to illuminate a finger
simultaneously. A mirror-reflected image is regarded as the                       uniformly.
“flipped” image taken by a virtual camera placed at a
different direction compared to the real one. Therefore, we

                                                                  53                                      http://sites.google.com/site/ijcsis/
                                                                                                          ISSN 1947-5500
                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                        Vol. 9, No. 3, March 2011

                                                                            can be used to fill "holes" of a size equal to or
                                                                            smaller than the structuring element. Used with
                                                                            binary images, where each pixel is either 1 or 0,
                                                                            dilation is similar to convolution. Over each pixel of
                                                                            the image, the origin of the structuring element is
                      Foreground separation                                 overlaid. If the image pixel is nonzero, each pixel of
                                                                            the structuring element is added to the result using
                                                                            the "or" operator. Used with greyscale images,
                           Normalization                                    which are always converted to byte type, the
                                                                            DILATE function is accomplished by taking the
                                                                            maximum of a set of sums. It can be used to
                           Gabor filtering                                  conveniently implement the neighbourhood
                                                                            maximum operator with the shape of the
                                                                            neighbourhood given by the structuring element.
                                                                            Used with greyscale images, which are always
                        Pyramidal method 
                                                                            converted to byte type, the ERODE function is
                                                                            accomplished by taking the minimum of a set of
                                                                            differences. It can be used to conveniently
                              Thinning                                      implement the neighbourhood minimum operator
                                                                            with the shape of the neighbourhood given by the
                                                                            structuring element.
                        Minutiae extraction                              B) Normalisation
                                                                            The process of removing the effects of the sensor
                                                                            noise and gray-level background due to finger
                                                                            pressure differences. The objective of this stage is
      Fig. 3: Overall flowchart of the proposed method                      decrease the dynamic range with gray scale between
                                                                            ridges and valleys of the image. Normalization
    3) Mirror: Two planar mirrors are positioned next to                    factor is calculated according to the mean and the
         the left and right side of the finger and the mirror               variance of the image. Each and every pixel in the
         size is determined to cover the maximum thumb                      fingerprint image has to be processed to find the
         size. To provide enough overlapping area between                   median value. The average value of all the pixels is
         frontal- and side-view images, the angles of the                   calculated i.e, the median value. By comparing the
         mirrors are determined 15 empirically. Also, the                   median value with the current pixel the replacement
         mirrors can be used as pegs to place a user’s finger               can be performed.
         firmly on the device.                                              Normalization facilitates have the subsequent
                                                                            processing steps.
                 III – PROPOSED METHOD                                      Let G (i, j) denote the normalized gray-level value at
In this section, we explain the Enhancement method for                      pixel (i, j). The normalized image is defined as
synthesizing an expanded fingerprint image from frontal- and                follows:
side-view images. The overall scheme of the method is
presented in Fig. 3 The method is mainly composed of six
stages (foreground separation, normalisation, Gabor filtering,
pyramidal method, thinning, minutiae extraction). In                                                                               (3)
foreground separation we will do the morphological
operation, in normalisation we pre-process the image etc.
     A) Foreground separation                                                     Where, M 0 and VAR0 denote the desired
         Using morphological operation we use the erosion                   mean and variance value, respectively.
         followed by dilation, this can be done up to required
                                                                            Most fingerprint images on a live-scan input device
         time. Mathematical morphology is a method of
         processing digital images on the basis of shape. A                 are usually of poor quality. The fingerprint image is
         discussion of this topic is beyond the scope of this               smoothed with an average or median filter.
         manual. A suggested reference is: Haralick,                     C) Gabor filtering
         Sternberg, and Zhuang, "Image Analysis Using                       A Gabor filter is a linear filter used in image
         Mathematical Morphology," IEEE Transactions on                     processing for edge detection. Frequency and
         Pattern Analysis and Machine Intelligence, Vol.                    orientation representations of Gabor filter are similar
         PAMI-9, No. 4, July, 1987, pp. 532-550. Much of                    to those of human visual system, and it has been
         this discussion is taken from that article. Briefly, the           found to be particularly appropriate for texture
         DILATE function returns the dilation of image by                   representation and discrimination. In the spatial
         the structuring element Structure. This operator is                domain, a 2D Gabor filter is a Gaussian kernel
         commonly known as "fill", "expand", or "grow." It                  function modulated by a sinusoidal plane wave. The

                                                                    54                             http://sites.google.com/site/ijcsis/
                                                                                                   ISSN 1947-5500
                                                    (IJCSIS) International Journal of Computer Science and Information Security,
                                                    Vol. 9, No. 3, March 2011

    Gabor filters are self-similar - all filters can be              Reduce the image size by a factor k        for three times.
    generated from one mother wavelet by dilation and                This is also outlined on the upper left hand side of Table
    rotation. Its impulse response is defined by a                   1. To create images containing only band limited signals
    harmonic function multiplied by a Gaussian                       of the original image, we expand the three images by
    function. Because of the multiplication-convolution              factor and subtract each of them from the next lower
    property (Convolution theorem), the Fourier                      level.
    transform of a Gabor filter's impulse response is the            E)   Thinning
    convolution of the Fourier transform of the                           The THIN function returns the "skeleton" of a bi-
    harmonic function and the Fourier transform of the                    level image. The skeleton of an object in an image is
    Gaussian function.                                                    a set of lines that reflect the shape of the object. The
     g ( x, y; λ,θ ,ϕ,σ , γ )                                             set of skeletal pixels can be considered to be the
                                                                          medial axis of the object. For a much more
                                                                          extensive discussion of skeletons and thinning
                                                                          algorithms, see Algorithms for Graphics and Image
                                                                          Processing, Theo Pavlidis, Computer Science Press,
                                                                          1982. The THIN function is adapted from Algorithm
                                                                          9.1 (the classical thinning algorithm).On input, the
                                                                          bi-level image is a rectangular array in which pixels
                                                    (4)                   that compose the object have a nonzero value. All
                                                                          other pixels are zero. The result is a byte type image
                                                                          in which skeletal pixels are set to 2 and all other
         Where x ' = x cos θ + y sin θ and                                pixels are zero.
                                                                     F)   Minutiae extraction
                                                                          A feature extractor finds the ridge endings and ridge
                   y = − x sin θ + y cos θ
                                                                          bifurcations from the input fingerprint images. If
                                                                          ridges can be perfectly located in an input
    In this equation, λ represents the wavelength of the                  fingerprint image, then minutiae extraction is just a
    cosine factor, θ represents the orientation of the                    trivial task of extracting singular points in a thinned
    normal to the parallel stripes of a Gabor function, φ                 ridge map. However, in practice, it is not always
    is the phase offset, σ is the sigma of the Gaussian                   possible to obtain a perfect ridge map. The
    envelope and γ is the spatial aspect ratio, and                       performance of currently available minutiae
    specifies the ellipticity of the support of the Gabor                 extraction algorithms depends heavily on the quality
    function.                                                             of the input fingerprint images. Due to a number of
                                                                          factors (aberrant formations of epidermal ridges of
D) Pyramidal method                                                       fingerprints, postnatal marks, occupational marks,
        Pyramid decomposition requires resizing                           problems with acquisition devices, etc.), fingerprint
   (scaling, or other geometric transformation). To                       images may not always have well-defined ridge
   create our Gaussian and Laplacian like pyramids, we                    structures. A reliable minutiae extraction algorithm
   define the reduce(I,K) and expand(I,K) operations,                     is critical to the performance of an automatic
   which decrease and increase an image in size by the                    identity authentication system using fingerprints.
   factor K, respectively. During reduce, the image is
   initially low-pass filtered to prevent aliasing using a
   Gaussian kernel.2. The latter’s standard deviation
   depends on the resizing factor, which here follows
   the lower bound approximation of the corresponding
   ideal low-pass filter                   . We initially
   reduce the original fingerprint image FP by a factor
   of              in order to exclude the highest
   frequencies. In a further step, we                                           Fig. 4: Types of Ridge Patterns

                       Table - 1
               Pyramidal building process

              a)   Pyramidal decomposition
       Gaussian-like               Laplacian-like

     G1=reduce(fp,k0)           L1=g1-expand(g2,k)
                                                                                     Fig. 5: Minutiae points
     G2=reduce(g1.k)            L2=g2-expand(g3,k)

                                                             55                                 http://sites.google.com/site/ijcsis/
                                                                                                ISSN 1947-5500
                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                           Vol. 9, No. 3, March 2011

        Minutiae are extracted from the thinned image by                 checking methods compare the foreground size of the fingers.
    using the Crossing Number algorithm.                                 Here foreground means the good quality regions of the finger
                                                                         print. The foreground size measures are tabulated as follows:


    Where Pi     0 or 1 in the 3*3 Neighbor of P

                     Characteristic of CN
               CN                         Character
                                                                                                     Fig. 8: Minutiae
                0                       Isolated point
                                                                                                          Table - 2
                2                         End point                       Average increasing rate of Foreground size in terms of each measurement

                4                     Bifurcation point                    Quality measurement                  Average increase rate of
                                                                                                                   foreground size
                                                                          Standard deviation [12]                      28.65%
For the experimental results we acquired 100 set of finger                     Coherence [13]                             33.72%
print images, each set contain frontal, left and right view
images. One of the used images set is shown in the Fig: 6 and                 Gradient – based                            30.81%
the enhanced image is also shown in the Fig: 7. The minutiae                    method [14]
extraction results also expressed in Fig: 8. The most definite
indicator of touchless image quality is the number of true               However we can expect that our enhanced image can be
minutiae additionally extracted.                                         making high performance when view difference image are
                                                                         matched. The Table-2 shows the result of our enhanced
                                                                                   V – CONCLUSIONS AND FUTURE WORK 
                                                                         This paper proposes a new method for touchless fingerprint
                                                                         sensing images. To get the better minutiae extraction, the
                                                                         three fingerprints (frontal, left, right) are enhanced using
                                                                         Gabor and pyramidal method. For experimental results, the
                                                                         enhanced fingerprints are having better enhanced ridges and
                                                                         the valleys. Also minutiae extraction is handled. The results
                                                                         are analysed and described in tables and graph format. In this
                        Fig. 6: Input images                             paper we limits the research work up to minutiae extraction,
                                                                         this research can be continued on mosaicing of the three
                                                                         enhanced images. Feature work can be done on the same
                                                                         concept. According to the result, it is concluded that the
                                                                         proposed system generate better enhancement on touchless
                                                                         fingerprint than the existing methods.

                                                                             [1]    D. Lee, K. Choi, and J. Kim, “A robust fingerprint matching
                                                                                   algorithm using local alignment,” in Proc. 16th Int. Conf. Pattern
                                                                                   Recognition, 2002, vol. 3, pp. 803–806.
                                                                             [2]   Hartwig Fronthaler, Klaus Kollreider, and Josef Bigun ,Local
                                                                                   Features for Enhancement and Minutiae Extraction in
                                                                                   Fingerprints, IEEE Transactions On Image Processing, VOL. 17,
                                                                                   NO. 3, MARCH 2008
                    Fig. 7: Enhanced images                                  [3]   Yi Chen1, Geppy Parziale2, Eva Diaz-Santana2, and Anil K Jain,
                                                                                   “3d Touchless Fingerprints: Compatibility With Legacy Rolled
Human experts prove that the more true minutiae extracted                          Images” Michigan State University Department of Computer
from the enhanced image. The touchless fingers are better                          Science and Engineering, 2006 Biometrics Symposium,
                                                                             [4]   TST Group Aug. 03, 2009 [Online]. Available: http://www.tst-
than the conventional touch based fingers, that conclusion                         biometrics.com
can be deviate from the results. The finger print quality

                                                                    56                                      http://sites.google.com/site/ijcsis/
                                                                                                            ISSN 1947-5500
                                                             (IJCSIS) International Journal of Computer Science and Information Security,
                                                             Vol. 9, No. 3, March 2011

[5]    Y. Song, C. Lee, and J. Kim, “A new scheme for touchless
       fingerprint recognition system,” in Proc. Int. Symp. Intelligent
       Signal Processing and Communication Systems, 2004, pp. 524–
[6]    Mitsubishi     Touchless     Fingerprint   Sensor     Aug.     03,
       2009[Online].Available: http://global.mitsubishielectric.com
[7]    N. K. Ratha and V. Govindaraju, Advances in Biometrics:
       Sensors, Algorithms and Systems. New York: Springer, 2008
[8]    TBS Touchless Fingerprint Imaging Aug. 03, 2009
       [Online].Available: http://www.tbsinc.com/
[9]    Y. Chen, G. Parziale, E. Diaz-Santana, and A. K. Jain, “3D
       touchless fingerprints: Compatibility with legacy rolled images,”
       in Proc. Biometric Consortium Conf., Baltimore, MD, 2006.
[10]   A. Fatehpuria, D. L. Lau, and L. G. Hassebrook, “Acquiring a 2-
       D rolled equivalent fingerprint image from a non-contact 3-D
       finger,” in SPIE Defense and Security Symp. Biometric
       Technology for Human Identification III, Orlando, FL, 2006, vol.
       6202, pp. 62020C-1–62020C-8.
[11]   Aug.          03,         2009        [Online].         Available:
       http://privacy.cs.cmu.edu/dataprivacy/          projects/handshot
[12]   L. Hong, Y.Wan, and A. K. Jain, “Fingerprint image
       enhancement: Algorithm and performance evaluation,” IEEE
       Trans. Pattern Anal. Mach Intell., vol. 20, no. 8, pp. 777–789,
       Aug. 1998.
[13]   E. Lim, X. Jiang, and W. Yau, “Fingerprint quality and validity
       analysis,”in IEEE Int. Conf. Image Processing (ICIP), Sep. 2002,
       vol. 1,pp. 469–472.
[14]   S. Lee, H. Choi, and J. Kim, “Fingerprint quality index using
       gradientcomponents,” IEEE Trans. Inf. Forensics Security, vol. 3,
       no.          4,          pp.792–800,          Dec.          2008. 

                                                                            57                          http://sites.google.com/site/ijcsis/
                                                                                                        ISSN 1947-5500