16

IEEE SIGNAL PROCESSING LETTERS, VOL. 11, NO. 1, JANUARY 2004

Fingerprint Enhancement in the Singular Point Area Sen Wang, Student Member, IEEE, and Yangsheng Wang

Abstract—Minutiae extraction for automatic fingerprint identification system is one of the most important steps. However, the performance of minutiae extraction relies heavily on an enhancement algorithm. There are many algorithms of enhancement that depend on the local orientation field of fingerprint. In the singular point area, because the local orientation changes very rapidly, we almost cannot make it accurate. So, the result of enhancement in the singular point is very bad. In this letter, we present a new method and design a new filter to enhance fingerprint in the singular point area. We distinguish the singular point area first. Then we design a new filter to enhancement this area. Finally, we give results and conclusion. Experimental results show a significant improvement of the fingerprint enhancement in the singular point area, and the time is required for our algorithm is reduced. Index Terms—Enhancement, filter, fingerprint, singular point.

I. INTRODUCTION

F

INGERPRINT images are direction-oriented patterns formed by ridges and valleys. In recent years, fingerprints are most widely used for personal identification. Most automatic systems for fingerprint comparison are based on minutiae matching. Structures of ridges in fingerprint images are not always well defined, and an enhancement algorithm that can improve the clarity of ridge structures is required. Therefore, enhancement is one of the most important steps in the automatic fingerprint identification system. The singular points—core and delta—are the most important global characteristics of a fingerprint (Fig. 1). The singular point area is defined as a region where the ridge curvature is higher than normal and where the direction of the ridge changes rapidly [1], [2]. In [3], the Gabor filter was presented to enhance fingerprint images, and the Gabor filter depends on the orientation and ridge frequency very much. Because the local orientation changes very rapidly in the singular point area, we almost cannot make it accurate. So, the result of enhancement in the singular point is very bad. Because the performance of enhancement in the singular point area is very poor, we must design a new algorithm to enhance this special region. Hence, there have been many methods attempt to solve this problem. In [4], a Fourier-domain filter along multiple orientations was

Manuscript received September 1, 2002; revised January 20, 2003. The associate editor coordinating the review of this manuscript and approving it for publication was Dr. John Apostolopoulos. S. Wang was with the National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, 100080 Beijing, China. He is now with the Department of Computer Science, State University of New York at Stony Brook, Stony Brook, NY 11794 USA (e-mail: [email protected]). Y. Wang is with the National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, 100080 Beijing, China (e-mail: [email protected]). Digital Object Identifier 10.1109/LSP.2003.819351

Fig. 1. Singular point (the core and delta point).

used by analyzing local Fourier transforms. Whereas, in [5], it was proposed to used a set of multiply oriented logical–linear operators for a directional filter and testing its consistency with the logical preconditions for applying the filter. A specially tailored isotropic diffusion scheme was present in [6], and it resembled a filter that varied between a directional filter in regions with well-defined orientation and an isotropic filter near the area without a dominant orientation. The approach we present, which consists of switching to isotropic smoothing whenever the ridge curvature is high, is similar to part of the work in [6], but the implementation is completely different. We distinguish the singular point area (an area with high ridge curvature) from the regions with a dominant orientation and then deign a new isotropic filter to enhance the image in this area, whereas in [6] a continuous transition between both cases is made. In this letter, we present in Section II an approach to preprocess the input original fingerprint images. After that, in Section III, we compute orientation fields and define the singular point area. Then, we design a new filter to enhance this region in Section IV. Finally, in Section V, we give some experimental results. II. PREPROCESSING Because there is noise in original fingerprint images, and fingerprint images may be in bad quality, we cannot identify the singular point area efficiently. In order to reduce the influence of noise, we preprocess the input fingerprint images. A. Normalization Before process the input fingerprint image, we normalize the image to constant mean and variance. Normalization is done to remove the effects of sensor noise and finger pressure difdenotes the gray value at pixel . and ference. are the estimated mean and variance of the input fingerprint image VAR VAR VAR

where

if

otherwise VAR and VAR are the desired mean and variance values.

1070-9908/04$20.00 © 2004 IEEE

WANG AND WANG: FINGERPRINT ENHANCEMENT IN THE SINGULAR POINT AREA

17

B. Smooth and Histogram Equalization After normalization, we use Gaussian smoothing to reduce the influence of noises, and we use histogram equalization to make the input fingerprint image look clear. III. COMPUTING DIRECTIONAL FIELDS AND DEFINING THE SINGULAR POINT AREA A. Computing Orientation Fields We use the following method, which is presented by Jain [7], to compute the orientation fields of fingerprints. We divide the input fingerprint image into blocks (8 8) as

Fig. 2. Region of the singular point area.

where are the gradients at each pixel. is the direction of the block . Finally, we use Gaussian smoothing operator to smooth the orientation fields in a neighborhood. B. Defining the Singular Point Area In a fingerprint image, the core point is the singular point that we try to detect, and we define the core point as the center point [8], [9]. The core point detection algorithm is described below. Step 1) Estimate and smooth the directional fields of the input fingerprint image. Step 2) In each block (8 8), we compute the Poincare index. The Poincare index is defined and computed as follows: Poincare if if otherwise

where

is the directional field of fingerprint. and are the coordinates of the blocks hat are in the closed curve with blocks. Step 3) If the Poincare index is 1/2, then this block is the core block. The center of this block is the core point. If the Poincare index is 1/2, then this block is the delta block. The center of this block is the delta point. If more than two cores or delta point are detected, go back to Step 1), using a larger smoothing parameter for the directional fields. After we detect the core and the delta point, we can select the singular point area. From the experimental results, we define the singular point area as a square around the core and the delta point, and its sides are 100 pixels (Fig. 2).

Fig. 3. (1), (3) Original fingerprint images. (2), (4) fast Fourier transform of (1), (3) respectively (by removing the dc component).

IV. FILTER DESIGN Fingerprints contain ridges and valleys separately, and these ridges flow almost parallel to each other. However, in the singular point area, this pattern is changed. The ridge directions change rapidly. The Fourier spectrum of a small region of a fingerprint image reveals two high peaks except the direct current (dc) component, but in the singular point area, the main energy is around the center, and the two peaks are not clear [10] (Fig. 3). From what is discussed above, we design a bandpass filter to enhance the singular point area. Fig. 4 shows a designed filter in the Fourier domain. The following bandpass filter is designed as

otherwise where and are the center frequency and bandwidth of and are the low-cut and high-cut values the filter. that suppress the effect of the low-frequency and high-frequency noise, respectively. First, we divide the singular point area into many blocks (32 32). Then, the enhancement of fingerprint in the singular point area is performed as in the following steps. Step 1) Transform the fingerprint image in the singular point area into the Fourier domain

Step 2) Filter the image by the designed filter

18

IEEE SIGNAL PROCESSING LETTERS, VOL. 11, NO. 1, JANUARY 2004

Fig. 4.

Designed filter in the Fourier domain with  = 5:0,  = 4:0.

Step 3) Transform the image in the Fourier domain back into the spatial domain

where and are the original image and an image filtered by the designed filter. and are the original image in the Fourier domain and an image filtered by the designed filter and are the two-diin the Fourier domain. mensional forward and inverse Fourier transformation, respectively.

Fig. 5. Example of the enhancement in the singular point area. (1)–(3) Original fingerprint images in the singular point area. (4)–(6) Results of our enhancement algorithm after binarization. (7)–(9) Results of the Fourier filter enhancement algorithm after binarization. Images are captured by Veridicom COMS sensor. TABLE I DIFFERENT AR OF THE TWO ALGORITHMS IN THE TWO DATABASES AFTER BINARIZATION, THINNING, AND MINUTIAE EXTRACTION

TABLE II DIFFERENCE OF THE TWO ALGORITHMS IN THE THREE FINGERPRINT IMAGES OF THE FIG. 5 AFTER BINARIZATION, THINNING, AND MINUTIAE EXTRACTION

V. EXPERIMENTAL RESULTS Our purpose of the fingerprint enhancement is to improve the clarity of ridges and valleys of the input images in the singular point area. This new algorithm of fingerprint enhancement is tested on two databases: one is the standard database NIST 27 [11]. This database contains 258 cases. Each case consists of a latent image and its matching tenprint image (referred to as the mate), and we only use the tenprint image in each case. The scanning resolutions and sizes of these grayscale images are 500 pixels per inch (ppi) and 800 768 pixels, respectively. The other database we used is a database that consists of 1500 fingerprint images of size 300 300 captured by a Veridicom COMS sensor. The filter we designed depends two parameters. One is a scale parameter related to the ridge frequency; the other is a bandwidth parameter. Since ridge frequency may vary significantly form one image to the other, it is to be expected that the ridge frequency has to be tuned to each image. For given resolution, the value of ridge frequency in a local neighborhood lies in a certain range, and we use the method in [12] to compute ridge frequency. By cutting down the valid frequency range, we avoid wrong estimation of frequency in blocks that do not form a well-defined frequency. We compare the results of our algorithm with the results of the Fourier filter [13]. The experimental results show that our algorithm is better than the other. An example of the enhancement in the singular point area is shown in Fig. 5.

We also can see from the results that our algorithm is more suitable to the minutiae extraction than the other filters. We define the accuracy rate (AR) of the minutiae extraction as follows: Mr Mm Ms AR Mt where Mr is the number of the true minutiae we extract from the singular point area by used minutiae extraction algorithms. Mm is the number of the missing minutiae. Ms is the number of the spurious minutiae, and Mt is the number of the true minutiae detected by fingerprint proficient. Table I shows the different ARs of the two algorithms in the two databases, and Table II shows the difference of the two algorithms in the three fingerprint images of Fig. 5. From the result, our algorithm is better than the other enhancement algorithm in improving the performance of the minutiae extraction. Thus, we can draw a conclusion that our enhancement algorithm does im-

WANG AND WANG: FINGERPRINT ENHANCEMENT IN THE SINGULAR POINT AREA

prove the quality of the fingerprint image in the singular point area and improve the accuracy of the minutiae extraction.

VI. SUMMARY AND CONCLUSION We have developed a new and fast fingerprint enhancement algorithm that can improve the performance of the minutiae extraction. The performance of our algorithm is tested on two databases, compared with the other enhancement algorithm and evaluated by using the accurate rate of the minutiae extraction. Experimental results show that our enhancement algorithm is capable of improving both the quality of fingerprint image and the accuracy of the minutiae extraction.

REFERENCES [1] V. S. Srinivasan and N. N. Murthy, “Detection of singular points in fingerprint images,” Pattern Recognit., vol. 25, no. 2, pp. 139–153, 1992. [2] J. Boer, A. Bazen, and S. Cerez, “Indexing fingerprint database based on multiple feature,” in Proc. ProRISC 2001 Workshop on Circuits, Systems and Singal Processing, 2001.

19

[3] L. Hong, Y. Wan, and A. Jain, “Fingerprint image enhancement: Algorithm and performance evaluation,” IEEE Trans. Pattern Anal. Machine Intell., vol. 20, pp. 777–789, Aug. 1998. [4] R. Pradenas, “Directional enhancement in the frequency domain of fingerprint images,” Proc. SPIE, vol. 2932, pp. 150–160, 1997. [5] L. A. Iverson, “Toward discrete geometric models for early vision,” Ph.D., McGill Univ., Montreal, QB, Canada, 1993. [6] A. Almansa and T. Lindeberg, “Fingerprint enhancement by shape-adaptation of scale-space operators with automatic scale-selection,” IEEE Trans. Image Processing, vol. 9, pp. 2027–2042, Dec. 2000. [7] A. Jain, L. Hong, and R. Bolle, “On-line fingerprint verification,” IEEE Trans. Pattern Anal. Machine Intell., vol. 19, pp. 302–314, Apr. 1997. [8] K. Karu and A. Jain, “Fingerprint classification,” Pattern Recognit., vol. 29, no. 3, pp. 389–404, 1996. [9] S. Wang, W. W. Zhang, and Y. S. Wang, “Fingerprint classification by directional fields,” in Proc. 4th IEEE Int. Conf. Multimodal Interface, Pittsburgh, PA, 2002, pp. 395–398. [10] T. Kamei and M. Mizoguchi, “Image filter design for fingerprint enhancement,” in Proc. ISCV, 1995, pp. 109–114. [11] M. D. Garris and R. M. McCabe. (2000) NIST special database 27: Fingerprint minutiae from latent and matching tenprint images. NIST, Gaithersburg, MD. [CD-ROM] NIST Tech. Rep. NISTIR 6534 [12] Z. M. Kovacs, R. Rovatti, and M. Frazzoni, “Fingerprint ridge distance computation methodologies,” Pattern Recognit., vol. 33, pp. 69–80, 2000. [13] B. G. Sherlock, D. M. Monro, and K. Millard, “Fingerprint enhancement by directional fourier filtering,” Proc. Inst. Electr. Eng., Vis. Image Signal Process., vol. 141, no. 2, pp. 87–94, 1994.