JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 9
A Secondary Fingerprint Enhancement for Identification System Raju Rajkumar and K Hemachandran Abstract- A reliable algorithm for the fingerprint image analysis is described. FFT is applied on a block, of size 32 X 32 pixel, of the fingerprint image for enhancement and after enhancement of each block, Gaussian filter is applied in each interconnection of the block. This Gaussian filter makes ridge smoothness and also reduce the “hairy” structure in the ridge when it is thinning. These experimental result show that the results are acceptable for feature extraction purpose. Index Terms- Real minutiae, Spurious minutiae, ridge thinning, primary enhancement, secondary enhancement.
-------------------------------------◊------------------------------------
1. Introduction Fingerprints are graphical ridge patterns present on human fingers and have been proposed as the most reliable human characteristics that can be used for people identification, due to their uniqueness and permanence[1]. The fingerprint features which are commonly used for identification are ridge ending and ridge bifurcation, which are usually called minutiae as shown in Fig. 1. Several factors like the presence of scars, variations of the pressure between the finger and the image acquisition sensor, the environmental conditions during the acquisition process, worn artifacts and so forth, can dramatically affect the quality of the acquired fingerprint image. Since minutiae depend on the fine details of the ridge pattern, their extraction can become notoriously difficult if the noise generated by the factors described above is not substantially reduced. The main goals of a fingerprint image enhancement algorithm are: (i) to reduce the noise present in the image and (ii) to detect the fingerprint ridge.
___________________________ • •
Raju Rajkumar is with the Department of Computer Science, Assam University, Silchar. K Hemachandran is with the Department of Computer Science, Assam University, Silchar.
(a)
(b)
Fig. 1 Two commonly used fingerprint minutiae bifurcation (b) Ridge ending.
(a) Ridge
A number of enhancement algorithms have been proposed in the literature[2, 3, 4, 5, 7, 8]. O’ Gorman and Nickerson [2] have proposed different filters for the fingerprint image enhancement and the k X k mask coefficients are generated, based on the local ridge orientation. Only three orientation directions are used. The four model parameters derived from ridge width (Wmax, Wmin), valley width (Ẃmax, Ẃmin), and the minimum radius of curvature are used to describe a fingerprint. It is assumed that the Wmax + Wmin = Ẃmax +Ẃmin . The mask is convoled with the input image. The enhanced image is binarized and post processed. Mehtre [3] computes the directional image, representing the local ridge direction, in a block of size 16 X 16 pixels. For this purpose, local gray level intensity variances along eight different directions are computed. The direction with the least variance is the desired least direction. A set of eight 7 X 7 convolution masks is applied to the input image for ridge enhancement. The fingerprint area is segmented from the background before applying standard locally
© 2011 JCSE http://sites.google.com/site/jcseuk/
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 10
adaptive thresholding and thinning operators. Features are obtained based on the computation of the connection number (CN) described in[2]. Sherlock et. al.[4] have studied the enhanced fingerprint images by using a directional Fourier filter. The direction of the filtering is decided by the local ridge orientation. A 32 X 32 window is used to obtain a projection of the pattern in 16 directions. The projection with the maximum variance is the desired ridge direction for the window. The result of the enhancement is compared with feature extraction techniques used in a system by the UK Home office. Hong et. al. [5] introduced a new fingerprint enhancement algorithm which decomposes the input fingerprint image into a set of filtered images. A set of band pass filters can efficiently remove the undesired noise and preserve the true ridge/valley structure. Gabor filters [6] have both frequency –selective and orientation-selective properties and have optimal joint resolution in both spatial and frequency domains. Therefore, it is beneficial to use Gabor filters as bandpass filters to remove the noise and preserve true ridge/valley structure. One of the heuristics to detect the spurious minutiae resulting from these crack, is based on the observation that these minutiae are anti-aligned and the region between them is brighter than the average brightness of the foreground region. Hong et. al.[7] improved the computationally expensive enhancement method [5], which makes it unsuitable for an online verification system. The authors have presented a fast enhancement algorithm to enhance adaptively, the ridge and furrow structures using both the local ridge orientation and local frequency information. Yang et. al.[8] modified the method proposed by Hong et. al. [7] by discarding the inaccurate prior assumption of sinusoidal plane wave, and making the parameter selection process independent of fingerprint image. Wuzhili [9] developed an enhancement algorithm, which applies a set of intermediate steps on the input image, and finally output the enhanced image. Two methods are adopted for image enhancement stage: one method is histogram equalization and the other is based on Fourier transform by using 32 X 32 pixel block and
binarize the image by using adaptive binarization, segmenting the background and foreground image by Morphological operations and then thinning.
2. Proposed Algorithm The proposed algorithm in this paper originates from Wuzhili’s algorithm[9], with the modification in Histogram equalization and an additional step of Gaussian filter after primary FFT enhancement. The Gaussian filter is implemented, to achieve a more smoothness and clearer image at the time of ridge thinning. These modifications make easy to extract fingerprint genuine feature. The proposed algorithm is shown in Fig 2. The performance of an Automatic Fingerprint Recognition System depends critically upon the quality of enhancement achieved. The fingerprint images, acquired from sensors or other media are not always assured with perfect quality. The enhancement methods are needed, to increase the contrast between ridges and furrows and for connecting the false broken points of ridges. This enhancement algorithm receives an input fingerprint image, applies a set of intermediate steps on the input image, and finally outputs an enhanced image.
NORMALIZATION PRIMARY ENHANCEMENT SECONDARY ENHANCEMENT
© 2011 JCSE http://sites.google.com/site/jcseuk/
BINARIZATION RIDGE DIRECTION ROI THINNING REMOVE H-BREAKS REMOVE SPUR Fig 2. Proposed algorithm
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 11
2.1 Normalization Normalization is done so that the gray level values lies within a given set of values. The fingerprint image is normalized to have a predefined mean and variance. The normalization is required as the fingerprint image usually has distorted levels of gray values among the ridges and furrows. This is a pixel-wise operation although it does not change the ridge and furrows structure. Let I(i,j) denotes the gray-level value at pixel (I,j), M and VAR denote the estimated mean and variance of I, respectively, and G(i,j) denote the normalized gray-level value at pixel (i,j).The normalized image is defined as follows[ 7]: VAR Ii, j M
VAR , . ,
, !"#
!"# , $
3(A) Original Image
… 1'
where MO and VARO are the desired mean and variance values, respectively. Normalization is achieved by histogram equalization. It increases the local contrast in an image. Thus the intensities can be distributed on the histogram. This allows for areas of lower local contrast to gain a higher contrast without affecting the global contrast. Histogram equalization accomplishes this by effectively spreading out the intensity values. This is shown in Fig. 3 ( B), for the original image of Fig. 3(A).
2.2 Enhancement 3(B) After Normalization
In our algorithm, there are two stages of fingerprint enhancement vis, primary and secondary.
2.2.1 Primary enhancement
Fig 3. Image enhancement by normalization.
according to the following equation: ,-. 2-.
The primary enhancement is done through Fourier transformation. Initially the fingerprint image is divided into small processing blocks (32 X 32 pixels) and the Fourier transformation is performed
(), * + . +. 4, 5 6 exp :2< 6 = /01 301
… … … . 2
for u = 0,1,2,….,31 and v = 0,1,2,….,31.
© 2011 JCSE http://sites.google.com/site/jcseuk/
)4 *5 ?@ >
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 12
In order to enhance a specific block by its dominant frequencies, the FFT of the block was multiplied by its magnitude a set of times. Where the magnitude of the original FFT = |F(u,v)|. The enhanced block is obtained according to
A4, 5 ( -. B(), * 6 |(), *|D E ………….(3) -1
where F (F(u,v)) is done by: 4, 5
1 )4 *5 + + ( ), * 6 exp :2< 6 = ?@ > > ,-. 2-. /01
301
………….… … 4
for x = 0, 1, 2, ..., 31 and y = 0, 1, 2, ..., 31.
The k in formula (3) is an experimentally determined constant, as k = 0.45. While having a higher "k" value improves the appearance of the ridges, by filling up small holes in ridges and having a too high "k" value can result in false joining of ridges. Thus a termination might become a bifurcation. The enhanced image after FFT has the improvements to connect some falsely broken points on ridges and to remove some spurious connections between ridges as shown in Fig 4(A).
Fig 4 (B) After Secondary enhancement
2.2.2 Secondary enhancement With the usual enhancement methods, the results of the fingerprint images are still found with certain spurious minutiae. Most of the enhancement filters have produced spurious minutiae[4, 9, 17]. Many researchers have attempted to remove this spurious effects [7, 18, 19]. One of the reasons for the “hairy” (spikes) structures which lead to spurious ridge bifurcations and endings is the ridge boundary aberrations which have an adverse impact on the skeleton, at the time of thinning. This effect is reduced by blurring the image, after the enhancement as shown in Fig 4(B). More over the mismatch or alignment of the blocks are removed by the blurring, where each block of the image is consider as sinusoidal plane wave [7] and from one block to another block the ride lines are not smooth. By applying Gaussian filter for blurring, the ridge becomes smooth. A two dimensional Gaussian function is given by [15]:
4, 5 G
Fig. 4. (A) After Primary enhancement
HI JKI ILI
-
……..(5)
Where σ is the standard deviation. A large value of σ produces a flatter curve and a small value leads to a
© 2011 JCSE http://sites.google.com/site/jcseuk/
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 13
“pointier” curve. In this study, the value of σ is 1.5 and its figure is shown in Fig 5.
furrows. After the operation, ridges in the fingerprint are highlighted with black color while furrows with white. A locally adaptive binarization method is reported in literature to binarize the fingerprint image[21]. Such a named method comes from the mechanism of transforming a pixel value to 1 when the value is larger than the mean intensity value of the current block (16x16) to which the pixel belongs. The fingerprint image with binarization is shown in Fig. 7 .
Fig 5. Two-dimensional Gaussian with σ = 1.5
As shown in Fig.6, all the clear blocks are perfect sinusoidal plane wave and the blurred one are not truly sinusoidal plane wave[8]. From one block to another block, there are many zig-zag structures in the ridge flow, because of the different angle of the ridge direction[7, 18]. The smoothness of the ridges are achieved by applying the Gaussian filter.
Fig. 7. Binarization
2.4 Ridge Direction The algorithm for the estimation of the direction of each block of the fingerprint image with W X W size (W is 16 pixel by default) is given below: I.
II.
Fig 6. 32 x 32 block pixel of 4 X 6.
2.3 Binarization The Fingerprint image binarization is to transform the 8-bit gray fingerprint image to a 1-bit image with 0 value for ridges and 1 value for
© 2011 JCSE http://sites.google.com/site/jcseuk/
The gradient values, along x-direction gx and y-direction gy , for each pixel of the block, are calculated using two Sobel filters. For all the pixels in each block, the least square approximation of the block direction is obtained using the formula [18 ]: tg2ß = 2 ∑ ∑ (gx*gy)/∑ ∑ (gx2-gy2)
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 14
The formula is easy to understand by regarding gradient values along xdirection and y-direction as cosine value and sine value. So the tangent value of the block direction is expressed as: tg2θ = 2sinθ cosθ /(cos2θ -sin2θ ). III.
After the estimation of each block direction, those blocks without significant information on ridges and furrows are discarded based on a threshold for the certain level E: E = {2 ∑ ∑ (gx*gy)+ ∑ ∑ (gx2-gy2)}/ W*W*∑ ∑ (gx2+gy2) The direction map is shown in Fig 8 (A). Fig 8 (B). Region of Interest
2.5 Region Of Interest Morphological operation OPEN and CLOSE are used for the determination of ROI(Region Of Interest). The algorithm[9] throws away the leftmost, rightmost, uppermost and bottommost blocks out of the bound so as to get the tightly bounded region by containing the bound and inner area, as shown in fig 8(B).
2.6 Thinning Ridge thinning is to eliminate the redundant pixels of the ridges till the ridge is just one pixel wide, as shown in Fig.9. No further removal of the pixels should be possible after completion of thinning process. The proposed algorithm used the built-in Morphological thinning function in MATLAB environment .
Fig. 8 (A) Ridge direction
Fig 9. Ridge Thinning
© 2011 JCSE http://sites.google.com/site/jcseuk/
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 15
The objective of a good thinning algorithm are: a) To obtain a thinned fingerprint image with a single pixel width and no discontinuities. . b) To eliminate the noise and singular pixels.
before minutiae extraction, the validation algorithm will eliminate the erroneous pixels while preserving the skeleton connectivity at the fork regions. For these purpose, the morphological operations on clean, hbreak and spur are used. The results after removing the erroneous pixels are shown in Fig. 11(A) and 11(B)
2.7 Remove H break and Spur Ideally, the width of the skeleton should be strictly one pixel. However, this is not always true. There are still some erroneous pixel locations, where the skeleton has a two-pixel width as shown in Fig 9. An erroneous pixel is defined as the one with more than two 4-connected neighbors[22]. These erroneous pixels exist in the fork regions where bifurcations should be detected, but they have connection number (CN) = 2 instead of CN>2.
Fig. 11(A) Remove H-break
Fig 10. Example of erroneous pixel and their and their validation (bold-italic 0 : deleted erroneous pixel; bold-italic 1 : preserved erroneous pixels that are changed to normal pixels).
The existence of erroneous pixels may a) destroy the integrity of spurious bridges and spurs, b) interchange the type of minutiae character, and c) miss the detection of true minutiae. Therefore,
© 2011 JCSE http://sites.google.com/site/jcseuk/
Fig. 11 (B) Remove Spur
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 16
3. Experimental Results The fingerprint image processing algorithm with secondary enhancement as described in section 2 has been implemented and tested on a standard Fingerprint Verification Competition(FVC) 2004 database [20] which contains hundreds of fingerprint images. This enhancement algorithm is executed in MATLAB 7.3.0. Table 1 shows the execution times on different fingerprint images in the database(DB). The first three column
(A1) Wuzhili’s Algorithm (8_8.tif of DB3)
are from DB1, the second three are from DB2 and the last are from DB3. DB1, DB2 and DB3 were collected by using different optical sensor [12]. The average execution time for an image due to our proposed algorithm is 1.9026 seconds and due to wuzilli’s algorithm is 1.7845 seconds. Even though the execution time is more, the proposed algorithm gives better enhancement than the wuzhili’s algorithm with many ridge gaps joint and remove many “hairy” structure in the thin ridge, as shown in Fig 12.
(A2) Proposed Algorithm(8_8.tif of DB3)
(B1) Wuzhili’s Algorithm (3_6.tif of DB2) (B2) Proposed Algorithm (3_6.tif of DB2) Fig. 12. The proposed algorithm shows improvement from wuzhili’s algorithm.
© 2011 JCSE http://sites.google.com/site/jcseuk/
JOURNAL OF COMPUTER SCIENCE AND ENGINEERING, VOLUME 6, ISSUE 1, MARCH 2011 17
FVC2004 Fingerprint Numbers
DB1
Proposed Algorithm Wuzilli’s Algorithm
DB2
DB3
101_1 .tif
101_6. tif
9_4.tif
1_1.tif
4_8.tif
5_4.tif
1_3.tif
2_3.tif
3_3.tif
1.3740
1.2878
2.567
2.013
1.9584
1.6843
2.0655
2.2331
1.9384
1.0327
0.9592
1.7236
1.4637
1.6396
2.7912
2.5285
1.5458
2.3758
Table1. Execution times(in sec ) for different fingerprint images for proposed algorithm and Wuzhili’s algorithm.
The additional step of Gaussian filter gives more smoothing at the time of thinning. This helps in removing H-breaks and spur.
[10]
[11]
4. Conclusions We have proposed an improved enhancement algorithm over wuzilli’s algorithm for fingerprint identification[9]. The implementation of Gaussian filter after Fast Fourier transformation enhancement gives little bit of blur to the fingerprint image. However, this contributes a uniform ridge thinning. The proposed algorithm gives good segmentation between foreground and background and gives better smoothness of the ridge.
[12] [13] [14] [15] [16] [17]
References [19] [1] [2] [3]
[4]
[5] [6] [7]
[8]
[9]
Henry C. Lee and R.E. Gaensslen, editors. Advance in fingerprint technology. Elsevier, New York, 1991. L.O ‘Gorman and J.V. Nickerson. An approach to fingerprint filter design. Pattern Recognition, 22(1): 29-38, 1989. B.M. Mehtre. Fingerprint image analysis for automatic identification. Machine Vision and Applications, 6:124139,1993. B.G. Sherlock, D.M. Monro, and K.Millard. Fingerprint enhancement by directional Fourier filtering. IEE Proc. Vis Image Signal Processing, 141(2): 87-94, April 1994. L.Hong, A.K Jain, S Pankanti, and Ruud Bolle, ”Fingerprint enhancement” IEEE 5:202-207,1996 D. Gabor, Theory of communication, Journal of IEE,92:429457, 1946. L.Hong, Y.Wan, and A.K.Jain, “Fingerprint image enhancement: algorithm and performance evaluation.," IEEE Transactions on Pattern Analysis and Machine Intelligence 20(8), pp. 777-789, 1998. J.Yang, L.Liu, T.Jiang and Y.Fan “A modified Gabor filter design method for fingerprint image enhancement”. Pattern Recognition letter, 24:1805-1817, August 2003. Wuzhili, “Fingerprint Recognition”, Thesis, Hong Kong, Baptist University, 2002.
[20]
[21] [22]
[23]
L.C. Jain, U. Halici, I. Hayashi, S.B. Lee, and S. Tsutsui, “Intelligent biometric techniques in fingerprint and face recognition”, The CRC Press, 1999. D. Maio, and D. Maltoni, “Direct gray-scale minutiae detection in fingerprints”, IEEE Transactions Pattern Analysis and Machine Intelligence, vol. 19(1), pp. 27-40, 1997. D.Maio, D. Maltoni, R. Cappelli, J.L. Wayman, A.K. Jain, “FVC2004: Fingerprint verification competition. D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, “Handbook of Fingerprint Recognition”, Springer, New York, NY, 2003. R.C Gonzales & R.E. Woods, “Digital Image Processing”. New York, USA: Addison-Wesley, 1993. A. Mcandrew, “Introduction to Digital Image Processing with Matlab”, Cengage Learning, Indian Edition, 2004. L. Hong, “Automatic Personal Identification Using Fingerprint”, PhD Thesis, 1998. L.O ‘Gorman and J.V. Nickerson, “Match Filter Design for Fingerprint Enhancement”. Proc. Int.Conf. on Acoustic Speech and Signal Processing, pp. 916-919. 1988. N.Ratha, S.Chen, and A.K. Jain, “Adaptive flow orientation based feature extraction in fingerprint images. Pattern Recognition, 28(11): 1657-1672, 1995. Almansa A. and Lindeberg T., “Fingerprint enhancement by shape adaptation of scale-space operators with automatic scale selection”, IEEE Transactions on image processing, vol. 9, no. 12, pp. 2027-2042, 2000. FVC2004 http://bias.csr.unibo.it/fvc2004 J.R. Parker,”Grey Level thresholding in badly illuminated images”, IEEE Trans. Pattern Anal. Mach. Intell. 13(8) (1991) 813-837. Feng Zhao, Xiaoou Tang,”Preprocessing and Postprocessing for skeleton-based fingerprint minutiae extraction”, Pattern Recognition 40(2007) 1270-1281.
Raju Rajkumar is a Ph.D Scholar in the Department of Computer Science, Assam University, Silchar. His Research interest are Biometrics and Pattern Recognition. He is a member of IAENG. Kattamanchi Hemachandran is a Professor in the Department of Computer Science, Assam University, SIlchar. Presently he is working in the fields of Image processing, Software Engineering and Distributed computing.
© 2011 JCSE http://sites.google.com/site/jcseuk/