THE INTERNATIONAL JOURNAL OF ENGINEERING AND INFORMATION TECHNOLOGY (IJEIT), VOL.6, NO.2,2020 611

Using Prewitt Operator as Gradient-Based Method for Fingerprint Singular Points Detection

Mohamed A. Sullabi Jamila H. AL-Montaser Libyan academy-Misurata Libyan academy-Misurata

Abstract—Singular Point (SP) is an essential feature in Fingerprints are classify into five classes [3], they are fingerprint images. It is considered as a fingerprint known as Henry classification; Arch, Tented Arch ,Left landmark. It is used for both fingerprint classification and Loop, Right Loop and Whorl. The prevalence rate for alignment in automatic fingerprint recognition systems. In these classes among people is 3.7%, 2.9%, 33.8%, 31.7% this paper the Prewitt operator is used as method for and 27.9% respectively, Each class have number of SPs, detecting of singular points in fingerprint image. The Prewitt operator is used as a gradient-based method of Whorl class has two core and two delta, Arch class does fingerprint image . Singular Points detection not have any SPs, Tented Arch, Left Loop and Right technique searches on the core point and the delta point in Loop have one core and one delta, SPs used to fingerprint classes. The experimental work has been classification fingerprint[4]. Figure 1 shows five evaluated using the four groups of the set (B) of the fingerprint classes. FVC2004 database that include DB1, DB2, DB3 and DB4. The experimental evaluation of the Prewitt operator showed good results, especially in good-quality fingerprint images. The maximum accuracy achieved was in DB4 is up to 95% for core point detection and 53.75% for delta point detection

Index Terms: Biometrics, Singular point, Prewitt operator, Orientation field estimation.

I. INTRODUCTION fingerprint is the most common biometric for A personal recognition. The biometric system identifies or authenticate a person using physical characteristics of a person like a fingerprint, DNA, hand, Figure 1. Five Fingerprint Classes [7] face, iris or behavioral characteristics like speech or signature. The fingerprint has been given great attention The paper is organized as follows: In section 2, to developing use to make it faster and more accurate previous studies on the SPs detection, in section 3, results in fingerprint recognition systems, because most proposed technique and process to detect the core and of the real time applications need less time to the delta, in section 4, experimental results and conclusion execution of the recognitions[1], this is available in will be in sections 5. fingerprint recognition. The fingerprint is a duplicate of a fingerprint epidermis II. LITERATURE SURVEY when a person touches a smooth surface[2], in actual, the ridges and valleys shape the fingerprint epidermis. Most previous studies have focused on core point The fingerprint divide into three levels: global, local and detection, only a few works in literature demonstrate Fine-detail [3] , which appear different features for every delta point detection. This section presents a review of level. Global level contains Singular Points (SPs), also literature on singular points detection. called singularity, consists of core and delta. The core Song and Elliott [6] they proposed a method for core defined as the most point on the inner most ridges and a point detection using an orientation map, they used a delta defined as the center point where three different color look-up table, composed of four-color regions. directions flows meet. is used to computing gradient in orientation field estimation, then the smoothing orientation field by Gaussian smoothing operator is done ـــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــ Received 11 Mar, 2020; revised 9 May, 2020; accepted 9 May, to obtain an orientation map. The core point is detected 2020. by searching a point placed in the lowest part of the Available online 10 May, 2020. www.ijeit.misuratau.edu.ly ISSN 2410-4256 Paper ID: EN118 167 M.Sullabi and J.AL-Montaser/Using Prewitt Operator as Gradient-Based Method for Fingerprint Singular Points Detection upper-part maximum edge region. The proposed method, figure 2. The methodology divided into three important used images form FVC 2004 database (set A), the steps. obtained detection rates are 95% for DB2 and 97% for DB3. Fingerprint Fingerprint Singular Points The algorithm introduced by [7] for optimal core point Enhancement Segmentation Detection detection, using improved segmentation and orientation. The algorithm is suitable for low quality fingerprint Figure 2. Stages of SPs Detection images. Preprocessing here is required to locate core Fingerprint enhancement is used to enhance the effect point correctly. The modified in segmentation technique of sensor and gray level background by make the level detects Region Of Interest (ROI) by computing the mean values lie within a given set of values. The step used to and standard deviation of gradient of the image. The enhance is called normalization [4]. Fingerprint results of segmentation proposed show the performance segmentation is used to extract the region of interest is better than the Mean and Variance based segmentation (ROI), mean and variance is used to segmentation [4]. technique. The algorithm is applied on fingerprint images Singular Points detection mainly involve core point and form FVC2004 database .The proposed technique of fine delta point detection. In this paper, singular point orientation field estimation gives good results in oily and detection uses edge detection by apply Prewitt operator in dry images. computed orientation field estimation algorithm based on In addition, [2]introduces an algorithm for singular gradient method. point detection based on the fingerprint orientation field Orientation field estimation is the technique to reliability, the algorithm starts by enhancing the determine the directions of ridges in fingerprint image, fingerprint image using the Short Time Fourier the approach used in this paper to estimate orientation is Transform analysis(STFT); followed by calculating the gradient based, Prewitt operator is used to compute orientation field. Sobel operator is used to compute gradient. gradient in orientation field, after the computation of the Prewitt operator will measure two components. The orientation field, reliability for orientation field is vertical edge component is calculated with kernel Gx and computed and locating the singular points. The proposed the horizontal edge component is calculated with kernel algorithm has been evaluated on FVC2002 database DB1, Gy. The masks of the Prewitt operator are the result for the proposed method is 2% locating the given below in figure 3: singular points with spurious (false SP), 1.5% missing locating the singular points (missed SP). According to[8], second derivate of Gaussian filter to calculated the gradient in orientation field estimation in detect core point detection, also Geometry of Region(GR) technique is achieved in the fine finding of core point. FVC2004 database DB1 (set A) was applied for experiments. Images in this study categorized as good, bad and low quality. For good quality images, 98 images Figure 3. Prewitt Operator on Horizontal and Vertical Directions [12] showed correct core point out of 100 images. The research [1], presents a mask that locates the core Matlab Language providing function, to compute point at the end of the discontinuous line appearing in the gradient, we used a Matlab expression as in equation (1) orientation map. The experiment use FVC2002 (set A) to implement gradient magnitude and gradient direction DB1 and DB2 and FVC2004 DB1 (set A) databases. using Prewitt's gradient operator: Proposed mask achieve accuracy of 96.66%, 91.06% and 87.5% respectively. Significantly, also[4] proposes [Gx, Gy] = imgradientxy (I, 'prewitt') (1) modified Poincare Index method for resolving the limitations of some existing fingerprint singular point After enhancement and segmentation steps. The steps detection algorithms like inaccurate detection, and for orientation field estimation is in the following [9,10],: elimination of forged detection in Poincare Index method, 1. We divide the fingerprint image into non- The experiment study on FVC2000 database DB1,a overlapping blocks of size W × W of the image, modified Poincare index method is suitable for image of W is set equal to 16. medium quality, but it is not good for poor quality 2. Compute the gradients based on Prewitt operator images. using equation (1). Although studies have been conducted by many 3. Compute the local orientation felid using authors, this problem is still needing more attention, so to equation (2). fill this literature gap of delta detection. The study in this paper is addresses singular points detection for both; core point and delta point detection.

III. THE PROPOSED TECHNIQUE FOR 4. Smoothing Orientation field estimation using SPS DETECTION equation (3)

The proposed technique of this study for singular points detection is following the methodology shows in

www.ijeit.misuratau.edu.ly ISSN 2410-4256 Paper ID: EN118 THE INTERNATIONAL JOURNAL OF ENGINEERING AND INFORMATION TECHNOLOGY (IJEIT), VOL.6, NO.2,2020 611 5. Before applying equation (3), we need to Table 1. SPs Detection and Not Detection in Datasets Detected convert orientation felid image into a continuous No Data Set SPs vector felid by following equations (4), (5): Correct Forged Detected Detection Detection (4) Core Point 69 5 7 DB1 And Delta Point 21 16 43

Core Point 55 24 1 DB2 Delta Point 10 45 25 Then Applying the Gaussian low-pass filter as Core Point 54 26 0 DB3 following equations (6), (7): Delta Point 17 62 1 Core Point 76 4 0 DB4 Delta Point 43 15 22

The correct detection describes the accurate results of Where and are the sets after filtering, the core point and delta point detection. In table 2, the is the Gaussian low-pass filter. high detection rates were recorded with core points, but The technique used for singular point detection is with delta points, the detection rates were low compared based on Poincare index [10]. to core points. The low delta rate is attributed to either the I. Let θ(x,y) denote the direction of the pixel (x,y) in forged detection or missing delta in the fingerprint image. an M×N fingerprint image. Figure 4 shows sample of correct detection of core point II. Compute Poincare index at each pixel of the and delta point. orientated image by equation (8). Table 2. Detection Rates of Datasets

SPs Detection DB1 DB2 DB3 DB4

Core Point 86.25% 68.75% 67.5% 95% Where Delta Point 26.25% 12.5% 21.25% 53.75%

Gives the direction of any pixel in the image. for PC(i,j), the core point is located in 0.5 and delta point is located in -0.5. Figure 4. Correct Detection of Core Point and Delta Point For example: If PC(i,j)= 0.5, then core point is present at the location Figure 5 shows correct detection rates for core point (i,j). and delta point. It is clear that DB4 has the high rate If PC(i,j)=-0.5, then delta point is present at the location among all datasets. (i,j). Forged detection describes sudden detection for SPs. which is not desirable for detection, it falls under IV. EXPERIMENTAL RESULTS detection SPs in false place or detects missing SPs in the fingerprint image as shown in figure 6. The experimental study of the singular point detection DB2 and DB3 recorded the highest percentage for technique was performed on FVC 2004 database (set B) forged detection. It is important to highlight the fact that [11], which is publicly available. The FVC 2004 consists DB1 and DB4 recorded the lowest percentage for forged of DB1, DB2, DB3, and DB4. Each dataset contains of detection, figure 7 shows the details about forged 10 different sets of fingerprints and each set consists of 8 detection. fingerprint samples of single fingerprint. The accuracy test of the extracted core and delta points based on visual inspections of the evaluation. The results of experiments include; correct detection, no detection, and forged detection of SPs. Table 1 shows SPs results.

www.ijeit.misuratau.edu.ly ISSN 2410-4256 Paper ID: EN118 169 M.Sullabi and J.AL-Montaser/Using Prewitt Operator as Gradient-Based Method for Fingerprint Singular Points Detection

V. CONCLUSION

%

%

3

.

%

95

100.00% % This paper evaluated Prewitt operator to compute

5

86

%

. 75 . gradient in orientation felid to locate the singular points

80.00% 75

67

.

68

(core and delta). The results of the experiments recorded %

60.00% 53

%

25 the best result with DB4 as shown in core point and delta

.

25 %

40.00% .

5 26

. point detection. 21

20.00% 12 However, the proposed technique shows some modest 0.00% results in delta point detection because of missing or DB1 DB2 DB3 DB4 forged detection of delta point. Prewitt operator achieves better results when using Core Delta gradient compute in orientation field estimation,

especially in DB4 where the core point detection is 95% Figure 5. Correct Detection Rates and delta point detection is 53.75%. Through the experiments conducted in this research, it was found that some of the fingerprint images in FVC2004 (set B) are low quality due to the loss of singular points in the fingerprint images. This conclusion is an important result in understanding the influence of some factors on image quality.

REFERENCES [1] G.A. Bahgat, A.H. Khalil, N.S. Abdel Kader and S. Mashali, "Fast and accurate algorithm for core point detection in fingerprint images", Egyptian Informatics Journal, vol. 14, pp. Figure 6. Forged Detection for Core Point and Delta Point. 15-25, 2013. [2] M.S. Khalil, D. Muhammad, M.K. Khurran and K. Alghathbar,

"Singular points detection using fingerprint orientation field

reliability", International Journal of Physical Sciences, vol. 5, %

% no. 4, pp. 352-357, April 2010.

25 50

. .

80% % [3] M. Galar, J. Derrac, D. Peralta, I. Triguero, D. Paternain, C.

%

56 77

50

. Lopez-Molina, S. Garcia, Jose. M. Benitez, M. Pagola, E. %

60% %

75

. %

32 Barrenechea, H. Bustince and F. Herrera, "A survey of

25

30

.

18 20

40%

6 fingerprint classification part II: experimental analysis and %

20% 5 ensemble proposal", Knowledge-Based Systems, vol. 81, pp. 98-116, January 2015. 0% [4] G.B. Iwasokun, and O.C. Akinyokun, "Fingerprint singular DB1 DB2 DB3 DB4 point detection based on modified poincare index method". International Journal of. Signal Process. Image Processing. Core Delta Pattern Recognition, vol. 7, no.5, pp. 259-272, 2014. [5] A.I. Awad and K. Baba, "Singular point detection for efficient fingerprint classification". International Journal of New Figure 7. Forged Detection Rates Computer Architectures and their Applications (IJNCAA), vol. 2, no. 1, pp. 1-7, 2012. [6] Y.C. Song and S.J. Elliott, "Orientation-and edge-map-based In no detection, it is failed to detect of core point or fingerprint core-point detection". Optical Engineering, vol. 47, delta point, figure 8 shows the rates of no detection. no. 3, pp. 037201, March 2008. Obviously, DB3 and DB4 present best result in core point [7] M.U. Akram, A. Tarig, S.Nasir and A. Khanam, "Core point detection, there are no instances of no detection. In other detection using improved segmentation and orientation", presented at IEEE/ACS International Conference on Computer hand, the highest rate for no detection was 53.75%, it Systems and Applications, 2008. recorded in DB1 for delta point. [8] S.A. Surlakar, A.K. Surlakar, and A. Handa, "Reliable Core Point Detection in Fingerprint". International Journal of Advanced Research in Computer Science and Electronics

Engineering (IJARCSEE), vol. 1, no. 7, pp. 61-64, 2012. %

[9] Z.H. Hadi and S.S. Mahdi, "Fingerprint classification based on

75

.

% orientation field", International Journal of Embedded Systems 53 60.00% %

20 and Applications, vol. 8, no. 4, December 2018.

. 50

. [10] J.H. AL-Montaser and M.A. Sullabi, "Rule-based technique 31

40.00%

27 for fingerprint classification using sobel operator in

% %

% orientation field", presented at The 3rd Annual Conference on

75

25

.

20.00% .

25

8

. %

% Theories and Applications of Basic and Biosciences. Faculty of

1

1 0 0 science, Misurata University, September 2019. 0.00% [11] FVC 2004 web site: http://bias.csr.unibo.it/fvc2004 [21 April DB1 DB2 DB3 DB4 2020]. Core Delta [12] R.C. Gonzales and R.E. Woods, Digital Image Processing, 2nd ed, Prentice Hall of India Ltd, pp.578, 2004.

Figure 8. No Detection Rates

www.ijeit.misuratau.edu.ly ISSN 2410-4256 Paper ID: EN118