FINGERPRINT IMAGE SEGMENTATION BASED ON BOUNDARY

VALUES

M. Usman Akram, Anam Tariq, Shahida Jabeen and Shoab A. Khan

Department of Computer Engineering, EME College, NUST, Rawalpindi, Pakistan

Keywords:

Segmentation, Mean, Variance, Coherence, Gradient, Gray-level values.

Abstract:

A critical step in automatic ﬁngerprint identiﬁcation system(AFIS) is the accurate segmentation of ﬁngerprint

images. The objective of ﬁngerprint segmentation is to extract the region of interest(ROI). We present a method

for ﬁngerprint segmentation based on boundary area gray-level values. We also present a modiﬁed traditional

gradient based segmentation technique. The enhanced segmentation technique is tested on FVC2004 database

and results show that our modiﬁed method gives better results in all cases.

1 INTRODUCTION

Fingerprint recognition is regarded as the most pop-

ular biometric technique for person identiﬁcation. A

ﬁngerprint is a pattern of parallel ridges and valleys

on the surface of ﬁngertip (Zhou and Gu, 2004). Most

Automatic Fingerprint Identiﬁcation systems(AFIS)

are based on local ridge features; ridge ending and

ridge bifurcation, known as minutiae(A.K Jain and

Boole, 1997). Segmentation of ﬁngerprint image is

the most important part of AFIS. Feature extraction

algorithms extract a lot of false features when ap-

plied to the noisy background area. The purpose of

segmentation is to remove the noisy area at the bor-

ders of ﬁngerprint image. It is especially important

for the reliable extraction of ﬁngerprint features. Fig-

ure 1 shows ﬁngerprint segmentation. There are two

Figure 1: (a) Original Image (b) Segmented image.

types of ﬁngerprint segmentation algorithms: unsu-

pervised and supervised. In Unsupervised algorithms,

block wise features such as local histogram of ridge

orientation (Mehtre and Chatterjee, 1989),(Mehtre,

1987), gray-level variance, magnitude of the gradient

in each image block (N.K.Ratha and A.K.Jain, 1995),

Gabor feature (F.Alonso-Fernandez, 2005),(A.Bazen

and S.Gerez, 2001) are extracted. Supervised method

ﬁrst extracts several features like coherence, aver-

age gray level, variance and Gabor response (E.Zhu,

2006), then a simple linear classiﬁer is chosen for

classiﬁcation. Figure 2 shows the sample ﬁngerprint

images from set B of FVC2004 database(FVC, 2004).

Figure 2: Fingerprint images from FVC2004 database.

There are by far many algorithms focusing on the

segmentation of ﬁngerprint image. A method based

on local certainty level of the orientation ﬁeld was de-

scribed in (A. K. Jain and Bolle, 1997). In (Maio and

Maltoni, 1997) the average gradient on each block is

computed which is expected to be high in the fore-

ground and low in the background. In (A.Bazen

and S.Gerez, 2001) gradient coherence, gray inten-

sity mean and variance are also used in segmenta-

tion. The segmentation technique presented in (Lin-

Lin Shen and Koo, 2001) is based on Gabor ﬁlters.

All previous methods have some problems in differ-

ent cases especially when it is difﬁcult to distinguish

between foreground and background. In this paper we

have proposed a method which uses traditional tech-

134

Usman Akram M., Tariq A., Jabeen S. and Khan S. (2008).

FINGERPRINT IMAGE SEGMENTATION BASED ON BOUNDARY VALUES.

In Proceedings of the Third International Conference on Computer Vision Theory and Applications, pages 134-138

DOI: 10.5220/0001089401340138

 SciTePress

niques based on mean, variance, gradient and ridge

orientation. This method takes decision on the basis

of boundary area gray-level values.

This paper is organized in ﬁve sections. Section 2

presents the traditional segmentation techniques and

shows where they are good and where they fail to seg-

ment the image accurately. It also contains our mod-

iﬁed gradient based method. Section 3 contains the

proposed method for ﬁngerprint segmentation. Ex-

periment results of our technique compared with other

techniques are discussed in section 4 followed by con-

clusion in section 5.

2 SEGMENTATION

TECHNIQUES

In AFIS, the processing of the surrounding back-

ground in ﬁngerprint image is not necessary and con-

sumes more processing time in all stages. Cutting or

cropping out the region that contains the ﬁngerprint

feature(ROI) minimizes the number of operations on

the ﬁngerprint image.

2.1 Mean and Variance based Method

Steps for this method are summarized as follows:

1. Divide the input image I(i,j) into non-overlapping

blocks with size w x w. in our case w = 8.

2. Compute the mean value M(I) for each block us-

ing equation (1)

M(I) =

w/2

∑

i=−w/2

w/2

∑

j=−w/2

I(i, j) (1)

3. Use the mean value computed in step 2 to com-

pute the standard deviation value std(I) from equa-

tion (2)

std(I) =

w/2

∑

i=−w/2

w/2

∑

j=−w/2

(I(i, j) − M(I))

(2)

4. Select a threshold value. If the std(I) is greater

than threshold value, the block is considered as

foreground otherwise it belongs to background.

Figure 3 shows the segmented images based on

mean and variance method. Images in 2nd row

show that this method is good only for light back-

grounds.

Figure 3: 1st Row: Fingerprint Images from FVC2004

database, 2nd Row: Mean and Variance Based Segmented

Images.

2.2 Modiﬁed Gradient based Method

We have modiﬁed the traditional gradient based

method and steps for our method are summerized as

follows:

1. Divide the input image I(i,j) into non-overlapping

blocks with size w x w. In our case w = 8.

2. Use histogram equalization to enhance the con-

trast between background and foreground.

3. Use a 3x3 median ﬁlter to reduce the noise in

background of the image(Lim and S.Jae, 1990).

4. Compute the gradients ∂

(i, j) and ∂

(i, j) at each

pixel (i,j)which is the center of the block.

5. Compute the mean values M

and M

for x and

y component of the gradient using equations (3)

and (4) respectively

w/2

∑

i=−w/2

w/2

∑

j=−w/2

∂

(i, j) (3)

w/2

∑

i=−w/2

w/2

∑

j=−w/2

∂

(i, j) (4)

6. Compute standard deviation for both M

and M

using equations (5) and (6)

std

w/2

∑

i=−w/2

w/2

∑

j=−w/2

(∂

(i, j) − M

(I))

(5)

std

w/2

∑

i=−w/2

w/2

∑

j=−w/2

(∂

(i, j) − M

(I))

(6)

7. Compute the gradient deviation using equation(7)

grddev = std

+ std

(7)

FINGERPRINT IMAGE SEGMENTATION BASED ON BOUNDARY VALUES

135

8. Select a threshold value. If grddev is greater than

threshold value, the block is considered as fore-

ground otherwise it belongs to background.

Figure 4 shows the segmented images based on

gradient based method. Images in 2nd row of ﬁg-

ure 4 show that this method is good for dark back-

ground.

Figure 4: 1st Row: Fingerprint Images from FVC2004

database, 2nd Row: Gradient Based Segmented Images.

2.3 Direction based Method

A ﬁngerprint consists of parallel line structures. The

coherence will be considerably higher in the fore-

ground than in the background. The steps for cal-

culating the coherence(Bazen and Gerez., 2000) are

summarized as follows

1. Divide the input image I(i,j) into non-overlapping

blocks with size w x w. In our case w = 8.

2. Use 3x3 sobel vertical and horizontal masks de-

ﬁned in equations (8) and (9) respectively to com-

pute the gradients∂

(i, j) and ∂

(i, j) and at each

pixel (i , j)which is the center of the block.

sobelHorizontal =





−1 0 1

2 0 2

−1 0 1





(8)

sobelVertical =





−1 −2 −1

0 0 0

1 2 1





(9)

3. Estimate the local orientation using equations

(10), (11) and (12).

(i, j) =

i+w/2

∑

u=i−w/2

j+w/2

∑

v= j−w/2

(∂

(u,v))(∂

(u,v))

(10)

(i, j) =

i+w/2

∑

u=i−w/2

j+w/2

∑

v= j−w/2

∂

(u,v)− ∂

(u,v) (11)

(i, j) =

i+w/2

∑

u=i−w/2

j+w/2

∑

v= j−w/2

(∂

(u,v) + ∂

(u,v))

(12)

4. Calculate background certainty and orientation

ﬁeld using equation (13)

coh =

(i, j) +V

(i, j))

∗ V

(13)

5. Select a threshold value empirically. If the coh is

greater than threshold value, the block is consid-

ered as foreground otherwise it belongs to back-

ground.

Figure 5 shows the segmented images based on

direction based method. Images in 2nd row show

that this method is good for light background and

it gives almost same results as mean and variance

based technique.

Figure 5: 1st Row: Fingerprint Images from FVC2004

database, 2nd Row: Direction Based Segmented Images.

3 ENHANCED SEGMENTATION

METHOD

In section 2, three methods are discussed and their

results are also shown in ﬁgures 3, 4 and 5. In this

section, we present a method which is summarized as

follows

1. Divide the input image I(i,j) into non-overlapping

blocks with size w x w. In our case w = 8.

2. Locate the blocks that contains the boundary of

the ﬁngerprint image.

3. Compute the gray-level values for those blocks.

4. Set a threshold T almost near to 240(whitish gray

color). if block pixels have gray value greater

than the threshold, use mean and variance based

method or direction based method. Otherwise use

gradient based method which is modiﬁed in this

paper and is discussed in section 2.

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

136

Figure 6: 1st Row: Fingerprint Images from FVC2004

database, 2nd Row: Enhanced Method Based Segmented

Images.

Table 1: Results and Comparison-I.

Accurately Accurately Poorly Poorly

Approaches Segmented Segmented Segmented Segmented

(Numbers) (%) (Numbers) (%)

Mean and

Variance 213 66.5 107 33.5

Gradient

Based 294 91.8 26 8.2

Direction

Based 227 71.0 93 29.0

Enhanced

Segmentation 303 94.6 17 5.4

Figure 6 shows the segmented images based on en-

hanced segmentation method. Images in 2nd row

show that this method is good for all quality images.

4 EXPERIMENTAL RESULTS

Our modiﬁcation is veriﬁed on FVC2004(FVC, 2004)

database(DB1-B, DB2-B, DB3-B and DB4-B). It

contains 320 images of different sizes, scanned with

500 dpi resolution. The technique is also tested on

low quality, dry and high pressure ﬁngerprint images.

The results are compared with other techniques and

they are summarized in table 1 and table 2.The de-

cision for accurate segmentation and poor segmen-

tation is based on human eye observation. Figure 7

shows the comparison of all techniques and the pro-

posed method uses the best result from the results of

methods discussed in section 2.

5 CONCLUSIONS

In this paper a new modiﬁed method for ﬁngerprint

segmentation is proposed. Firstly three old meth-

ods are discussed and also their results are analyzed

for different quality ﬁngerprint images. Then a new

method is presented for accurate ﬁngerprint segmen-

tation. Finally experimental results are compared

with that obtained by traditional methods and our ﬁn-

Table 2: Results and Comparison-I.I.

Mean and Gradient Direction Enhanced

Quality Variance Based Based Segmenteion

of image (%) (%) (%) (%)

White

Background 97 87 95 97

Grayish

Background 87 89.3 91 91

Dark

Background 6 92 12 92

Noisy

Background 42 91.0 57 91

Figure 7: 1st Column: Fingerprint Images from FVC2004

database, 2nd column: Mean and Variance Based Seg-

mented Images, 3rd column: Gradient Based Segmented

Images, 4th column: Direction Based Segmented Images,

5th column: Enhanced Segmentation Based Segmented Im-

ages.

gerprint segmentation technique gives better results

for all kind of ﬁngerprint images.

REFERENCES

A. K. Jain, L. Hong, S. P. and Bolle, R. (Sept. 1997). An

identity authentication system using ﬁngerprints. In

Proc. IEEE, vol. 85, pp. 1365-1388.

A.Bazen and S.Gerez (2001). Segmentation of ﬁngerprint

images. In Proc. Workshop on Circuits Systems and

Signal Processing (ProRISC 2001), pp. 276-280.

A.K Jain, H. L. and Boole, R. (1997). One-line ﬁnger-

print veriﬁcation. In IEEE Trans. PAMI, Vol.19,No.4,

pp.302-314.

FINGERPRINT IMAGE SEGMENTATION BASED ON BOUNDARY VALUES

137

Bazen, A. and Gerez., S. (2000). Directional ﬁeld com-

putation for ﬁngerprints based on the principal com-

ponent analysis of local gradients. In ProRISC 2000

Workshop on Circuits, Systems and Signal Processing,

Veldhoven, The Netherlands.

E.Zhu, Yin, J. a. (2006). A systematic method for ﬁnger-

print ridge orientation estimation and image segmen-

tation. In Pattern Recognition,Vol.39, No.8, pp.1452-

1472.

F.Alonso-Fernandez, J.Fierrez-Aguilar, J.-G. (2005). An

enhanced gabor ﬁlter-based segmentation algorithm

for ﬁngerprint recognition systems. In Proceedings of

the 4th International Symposium on Image and Signal

Processing and Analysis(ISPA 2005). pp. 239-244.

FVC (2004). Finger print veriﬁcation contest 2004. In

Available at (http://bias.csr.unibo.it/fvc2004.html).

Lim and S.Jae (1990). Two-Dimensional Signal Image Pro-

cessing.

LinLin Shen, A. K. and Koo, W. (2001). Quality measures

of ﬁngerprint images. In 3rd Audio and Video-Based

Person Authentication, AVBPA ,pp. 266-271.

Maio, D. and Maltoni, D. (Jan 1997). Direct gray-scale

minutiae detection in ﬁngerprints. In IEEE Trans. on

Pattern Analysis and Machine Inteligence, vol. 19, pp.

27-40.

Mehtre and Chatterjee (1989). Segmentation of ﬁngerprint

images - a composite method. In Pattern Recognition,

Vol.22, No 4, pp.381-385.

Mehtre, B.M., M. N. K.-S. C. (1987). Segmentation of ﬁn-

gerprint images using the directional image. In Pattern

Recognition, Vol 20, No.4, pp.429-435.

N.K.Ratha, S. and A.K.Jain (1995). Adaptive ﬂow

orientation-based feature extraction in ﬁngerprint im-

ages. In Pattern Recognition, Vol.28, No.11, pp.1657-

1672.

Zhou, J. and Gu, J. (JUNE 2004). A model-based method

for the computation of ﬁngerprints orientation ﬁeld.

In IEEE Trans. on Image Processing, VOL. 13, No. 6,

pp.821-835.

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

138