Fingerprint Scholar (ECE) NITTTR , Chandigarh [email protected] Rohini ME

Fingerprint Enhancement Using Frequency Domain Filter

Anavil Mishra

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

ME Scholar (ECE)

NITTTR , Chandigarh

[email protected]

                                                                                                  

     Rohini

ME Scholar (ECE)

NITTTR , Chandigarh

[email protected]

 

 

In the present paper we will discuss about
fingerprint image enhancement for fingerprint recognition. Many experiments have been
done in past for fingerprint image enhancement. Many new procedures &
advances have developed gradually which has improved the poor quality
fingerprint images in a remarkable way. Therefore, in the given paper, an
assessment of these procedures & advances are done with its rationale
points to validate the results. Based on the given procedure, a consolidated
solution for fingerprint identification is evolved for affirmation. For the plan of action some methods at the time of coding &
calculations are initiated to improve & upgrade the working of this
fingerprint recognition system. The results are proposed only after conducting
multiple experiments on variety of fingerprint images

 

Keywords: Biometrics, Image processing, Fast Fourier transform, Gaussian
Band pass Filter, signal
processing, image analysis, feature extraction, fingerprint image enhancement,
minutia matching.

1.    
Introduction

BIOMETRICS is one of the most reliable solutions to determine identity
of an individual by identifying their physiological or behavioral
characteristics. Some of the commonly used physiological characteristics and
behavioral for biometric recognition are face, hand & fingerprints, iris,
retina, DNA, signature, voice etc. Automated fingerprint identification system
(AFIS) is the most commonly used technique implemented across the globe. It has
therefore, now become most important in security field because of which it has
interested many researchers to continue doing their research on it. Every human
being have different patterns of crest (ridges) & troughs (valleys). The
image which is shown in Fig.1 is showing the surface of the fingerprint which
is taken where the dark lines are showing the ridges and the light lines in the
fingerprint image is showing the valley.

 

 

 

Fig. 1.  Ridge and Valley Structure 1

 

 Mainly, three standards about fingerprint are
mentioned below:

  

1.1
Unalterable: Fingerprint patterns are permanent and never alter during the
growth of human body.

1.2.
Unique: Every human being has unique fingerprint ridges & are
never identical in two different people.

1.3.
Grading: Grading of fingerprint patterns is limited. This makes the
logical grading of fingerprint pattern easier.

 

Images of fingerprint which are
obtained from people for identification purposes are not often clear and
sometimes quality is also not perfect. Different factors like scars, sometimes moisture
which is present in scanner and there are also many more reasons which are
responsible for affecting the quality of an image which can lead to failure or
maybe there are chances of false extraction of minutiae.

 

Hence, it has become necessary to
adapt techniques for image enhancement to reduce noise present in it and also
to improve statement of ridges against valleys. This paper contains, various
fingerprint image enhancement techniques and these are discussed. The outcome result
of these techniques are shown in the given Table 1 and results of these
techniques are also discussed in Section 5. This further propose the fingerprint
image enhancement in two different domains that are spatial domain &
frequency domain which  are discussed. Clarity
of fingerprint image and contrast of different fingerprint images are improved
by using technique that is the spatial domain technique and then Gaussian
Bandpass filter is used which is tuned to a
fixed range of frequencies to transform them to the different fourier image.  Resultants are experimented on original
samples of fingerprints by using filters to further analyze the frequency
domain which is generated directly.

 

2.    
 PROPOSED WORK
& BACKGROUND

In this section, to improve the
clarity of ridges & valleys of the fingerprint impression some calculations
& methods are initiated that have been initiated to improve the picture
clarity of ridges and valleys. In this Frequency Domain Filter is used which
allows the fingerprint images to combine with the large size filters
efficiently.

 

For the improvement of image
Gabor filters have been used as initiated in 7, 8, 9, 10. To use Gabor
filters,  four different parameters are
used that are local ridge orientation, local ridge frequency and the standard
deviation of the Gaussian envelop along the x-axis and y-axis (?x and ?y) must be
selected specifically to avoid false ridges & valleys. For other
fingerprint enhancement techniques the methodology has been explained in
Section 3 & the results of the methodology implemented is given in Section
5.

 

3.     FINGERPRINT IMAGE
ENHANCEMENT TECHNIQUES

 

The following techniques have been used to analyse
& enhance the fingerprint images:

 

3.1.  Short Time Fourier
Transform (STFT) Analysis:

In 2 3, as a
new technique STFT analysis has been introduced for fingerprint magnification.
Fig.2 below shows the audit of the initiated methodology.

 

 

 

 

 

  

 

 

 

Original
image                                                                                                         
                  Enhanced Image

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Fig. 2. Initiated methodology in 2

 

Based
on the contextual filtering a new fingerprint image enhancement algorithm in
the Fourier domain is presented. In Fig. 1, input image is divided into
projected windows during STFT analysis. It is presumed that the image within
this small window is soaked and can be modelled as a surface wave. The rough
calculations of ridge frequency and ridge orientation are then acquired.  For the computation of angular coherence the
energy map yielded from STFT analysis is used. This coherence image is used to
adapt to the angular bandwidth. Then, the resultant contextual information is
used to filter each window on the Fourier domain.

 

 

 

 

 

table

 

 

The
above calculated table can be summarised as under:

It
has two stages. The first stage is composed of STFT analysis and second stage
is composed of contextual filtering. The STFT stage furnishes the three images
-ridge orientation, ridge frequency and the block energy. These are further
utilised to calculate the region mask. This is almost identical to the
technique which is initiated in 4. Hence, all the intrinsic images which are
required to perform contextual filtering are produced out of the analysis phase.

 

3.2.   Pyramid-based Filtering

This technique uses the
familiar methods for improvement and for drawing particulars as initiated in
5. Two types of symmetries namely parabolic and linear are used to prototype
as an example & draw the confined structure in a fingerprint image &
these symmetries can be roughly calculated by separate filtering of the
orientation tensor. In the improvement stage, a Laplacian-like image pyramid is
used to fragment the original fingerprint image into sub bands correlating to
different spatial scales.The frequency directions originate from the frequency
structures where surround smoothing is done on the pyramid at different levels.
In Fig. 4, the reduction (I, k) and expansion (I, k) function are defined to
create the Gaussian & Laplacian pyramid. The function is shown below in
Fig. 4.

 

 

 

                                                     
a) Pyramid Decomposition PD

 

a)      
Gaussian-Like

Laplacian-Like

g1 = reduce (fp, ko);
g2 = reduce (g1,k);
g3 = reduce (g2,k);
g4 = reduce (g3,k);

l1 = g1- expand(g2,k);
l2 = g2-expand(g3,k);
l3 = g3- expand(g4,k)

 

b)      
Reconstruction R

fp = expand (……ko);
                            
expand (l3,k) +l2

 

Fig. 4.  Pyramid building process
in 5

 

Later
to detect the accurate position & direction of the minutia the parabolic
symmetry is added to the local fingerprint model. The parabolic symmetry is
further reduced if the linear symmetry is high whereas it is maintained vice
versa. All these steps are followed in compliance with extracting the finer
details.

3.3.  Curved Gabor Filters

The
Gabor Filters play a crucial role in the improvement of different types of
images & in the process of drawing Gabor features. In 6, the curved Gabor
filters were introduced purposely to upgrade the curved structures in noisy
images. The choice of filter parameters increases the smoothing power of the
magnified images without fabrication. In several
operations of image processing and pattern realization, Gabor functions 8, in
the form of Gabor filters 9 and Gabor wavelets 10 were enforced for various
functions

 

In
6, it is observed that curved Gabor filter is applied to the curved ridge
& valley structure to enhance the poor quality fingerprint images.
Initially, two orientation field estimation methods were combined to obtain
more sturdy estimations for fingerprint images which are extremely noisy in
nature.

 

Curved
Gabor filters allow the use of large curved regions without creating false
features. It is therefore a testified & proven fact that the curved Gabor Filters draw
better results in comparison to the traditional Gabor Filters.

 

 

 

Fig. 5. (Left): Original
Fingerprint Image, (Middle): Image drawn after the use of Traditional Gabor
filter, (Right): Image drawn after the use of Curved Gabor filter.

 

3.4.  Fast
Fourier Transform and Gaussian Bandpass Filters

Have been used to magnify
the fingerprint images.

 

3.4.1.  The subdivided image is further divided into
non converging blocks of size W x W as presented in 7, The total time taken
to calculate FFT is much faster when the block sizes are power of two8.Thus,
16 X 16 is suggested to use for computation of value.

3.4.2.  Further for every block, 2D Discrete Fast
Fourier Transform is suggested to apply based on 1. The value F(0,0) of the
DFT is called the DC coefficient. When u=v=0 is substituted in the definition
given in 1, then

 

F(0,0) = ? M-1 ? M-1  f(x,y) exp(0 ) = ? M-1 ? M-1  f(x,y) 
                                                                                        (2)                                                                                                       

                 X=0   y=0  
                                 x=0   y=0

 

The Fourier
transform is complex. If the real components are represented by R(u,v) and
Imaginary components are represented by I (u,v) then the Fourier range will be
defined as:

 

 =        

 

 

To envision, it is appropriate to have
zero frequency element DC in the center of the matrix 8. In comparison to
other values DC coefficient is larger. Fourier transform Fingerprint image is
upgraded by a log transformation as in 8 & is shown in fig. 6

 

 

 

Fig. 6. Spectrum of the image block

 

3.4.3.  In Fig. 7 extent of frequencies in Bandpass
filter is shown. The cut-off frequency values for lower & higher limit are
taken as 130 and 225 respectively for sharpness. Multiply the FFT spectrum of
the block by the filter H (u,v)

 

     G = H ? F

3.4.4.Counter FFT transform is used to recover the
filtered image.

 

3.4.5.On repeating the above mentioned process the
enhanced image is obtained after the initial calculations are applied as shown
in Fig. 8

 

 

Fig. 7. Image
obtained after using Gaussian Bandpass Filter

 

 

 

Fig. 8.
Magnified image after using Gaussian Bandpass Filter

 

 

4.    
RESULT:

4.1.  FOR
STFT , PYRAMID BASED FILTERING  &
CURVED GABOR FILTERING

 

 In the given Table 1 the outcome of validity
of techniques used for fingerprint image magnification are given, based on the
EER’s percentage from the review papers mentioned above.

 

TABLE 1. RESULTS OF
FINGERPRINT IMAGE ENHANCEMENT TECHNIQUES

 

                                                                                                

 

Author

Techniques

Result/Accuracy

Bhowmik,etal. 1

DFT and Histogram Equilization

Quality of fingerprint image greatly increased and
more accurate minutiae extraction

Chikkerur,
Cartwright,
Govindaraju 2

Short Time Fourier Transform (STFT) Analysis

% of EER (FVC2004)
         DB1 =
19.1
         DB2 =
11.9
         DBE =
7.6
         DB4 =
10.9

 

 

Table 1 (Continued)

 

Fronthaler
Kollreider, Bigun 5

Pyramid- based filtering

% of EER (FVC2004)
         DB1 =
12.0
         DB2 =
8.2
         DBE =
7.6
         DB4 =
7.0

Gottschlich 7

Curved Gabour Filters

% of EER (FVC2004)
         DB1 =
9.7
         DB2 =
6.3
         DBE =
5.1
         DB4 =
6.5

 

 

EER = Equal Error
Rate, FAR = False Acceptance Rate, FRR = False Rejection Rate

 

(FROM THIS TABLE
REMOVE THE FOLLOWING HEADINGS AS I HAVE NOT INCLUDED THEM IN THIS REVIEW PAPER
– 1,5,7)

EER
value given in the above table shows that the segments of both FAR & FRR
are equal. The lower the EER value, higher the authenticity of the biometric
system. The functioning of curved Gabor filters 6 depends on the nature &
authenticity of the orientation fields (OF) and RF approximation.

 

4.2.  RESULT
FOR FAST FOURIER TRANSFORM & GAUSSIAN BANDPASS FILTERS WILL CONCLUDE THE
ANSWER

 

Assessments
were conducted on as many as 138 Fingerprint images gathered from 46 different
people from all class & categories using an Optical Fingerprint Reader. The
size of each image is 480 x 320. Highpass and Bandpass frequency domain filters
were used to affirm the best technique used in the process of fingerprint
improvement.

The
following guidelines & framework are investigate the methodology discussed
above.

 

4.2.1. PSNR: Peak
Signal-to-noise ratio

The
Peak Signal to Noise Ratio (PSNR) produce a degree

of
peak error in decibels between two images. This ratio is often used as a
quality measurement between the original image and a compressed or
reconstructed image 9. The technique with a higher PSNR produce better
results. PSNR is expressed as:

 

PSNR = 10log10                                                                                                                                           

 

Here
R is the ultimate desirable pixel value in the input image. The value of R is
taken as 255 for an 8-bit unspecified total data type gray scale image. The MSE
produces the aggregate squared error between the recreated & the pioneer
image. Here, the lower value of MSE shows lower error in the reformation of an
image. The mean-squared error is given as:

 

 

 =    I(I , j) – K ( I , j)2                                                                                  

 

where,
M & N are the number of rows and columns in the suggested images
subsequently. Image I is to be investigated & image K will work as the
relating image.

 

4.2.2.  Calculation
Time:

This measure is used to guage the
total time taken to magnify the image & can be ascertained by using cputime
command of Matlab.

 

Table-II
shows the conclusions drawn empirically by implementing frequency domain
filters on the sample Fingerprint images using Matlab.

 

TABLE II

EXPERIMENTAL
RESULTS

 

    

Enhancement Technique

Peak Signal to Noise ratio

Computation Time

Gaussian High pass Filter

7.4657db

0.8892 secs

Gaussian High pass Filter

26.4865db

0.8644 secs

 

 

5.    
 CONCLUSION:

 

The realization of a
fingerprint feature derivation & coordination calculations largely rely on
the quality of fingerprint images used for desired results. There have been
constant efforts made in enhancing the quality of fingerprint images as shown
in different studies. The oriented diffusion filter & curved Gabor filter
process give preferred results.

In the latter
half of the paper the other methods & techniques used for fingerprint
enhancement such as, Gaussian Bandpass filter is used in the frequency domain.
Here, with the use of PSNR the filtered image is capable of reducing the noise
& smoothening the holes & small breaks in a ridge in less time frame
required for calculation. The Fourier spectrum can be rotated undeviatingly
with the developed Bandpass Filter tuned to a limited range of cut off frequencies.
This work has concentrated on the circularly proportionate filters that are
stated as functions of the distance from the center of the filters. The
futuristic opportunity of this task can be to investigate more frequency domain
filters to accomplish improved outcomes.

 

6.    
REFERENCES:

 

 1 Davide Maltoni, Dario Maio, Anil K. Jain,
Salil Prabhakar, HandBook of Fingerprint Recognition, Second Edition,
Springer-Verlag (2009).

 

 2 Shaikh Mohammedsayeemuddin, Sima K
Gonsai and Dharmesh Vandra, “Efficient fingerprint image enhancement algorithm
based on Gabor filter”, International Journal of Research in Engineering and
Technology, Vol.3, pp. 809-813, 2014.

3
Yuan Mei, Bo Zhao, Yu Zhou, Songcan Chen, “Orthogonal curvedline Gabor filter
for fast fingerprint enhancement”, Electronics Letters, Biometrics Compendium,
IEEE, Vol.50, Issue 3, pp.175 – 177, 2014.

4
Miao-li Wen, Yan Liang, Quan Pan, Hong-cai Zhang, “A Gabor filter based
fingerprint enhancement algorithm in Wavelet domain”, IEEE International Symposium
on Communications and Information Technology, Vol. 2, pp.1468- 1471,2005

5
L. Hong, Y. Wan, and A. Jain, “Fingerprint image enhancement: Algorithm and
Performance Evaluation”, IEEE Transactions on Pattern Analysis and Machine
Intelligence, vol. 20, pp. 777-789, 1998

6
Miao-li Wen, Yan Liang, Quan Pan, and Hong-cai Zhang, “A Gabor filter based
fingerprint enhancement algorithm in Wavelet domain”, In proceeding of IEEE
International Symposium on Communications and Information Technology, Vol
2,pp.1421-1424, 2005

 7
S. Chikkerur, A.N. Cartwright and V. Govindaraju, “Fingerprint Enhancement
Using STFT Analysis”, Pattern Recognition, Vol. 40, No. 1, pp. 198-211, 2007.
(Journal)

 

8
S. Chikkerur, A.N. Cartwright and V. Govindaraju, “Fingerprint image
enhancement using STFT analysis,”Third International Conference on Advances in
Pattern Recognition, ICAPR 2005, Bath, UK, Proceedings, Part II, pp. 20–29,
2005. (Conference Proceedings)

 

9
L. Hong, Y. Wan and A. Jain, “Fingerprint Image Enhancement: Algorithm and
Performance Evaluation”, IEEE Trans. On Pattern Analysis and Machine
Intelligence, Vol. 20, No. 8,pp. 777-789, 1998. (IEEE Transactions)

 

10 H. Fronthaler, K.
Kollreider and J. Bigun, “Local Features for 
Enhancement and Minutiae Extraction in Fingerprints”, IEEE Trans. Image
Processing, Vol. 17, No. 3, pp. 354-363, 2008. (IEEE Transactions)

 

11
C. Gottschlich, “Curved Gabor Filters for Fingerprint Image Enhancement”, IEEE
Trans. On Image Processing, Vol. 21, No. 4, pp. 2220-2227, April 2012. (IEEE
Transactions)

12
D. Gabor, “Theory of Communication. Part 3: Frequency Compression and
Expansion”, Journal of the Institution of Electrical Engineers, Vol. 93, No.
26, pp. 429-441, 1946. (Journal)

13
D. Gabor, “Information Theory in Electron Microscopy”, Laboratory
Investigation, Vol.14, No. 6, pp. 801-807, 1965. (Article)

14
T.S. Lee, “Image Representation Using 2D Gabor Wavelets”, IEEE Trans. on
Pattern Analysis and Machine Intelligence, Vol. 18, No. 10, pp. 959-971, 1996.
(IEEE Transactions)

 

 15 Kumud Arora, Poonam Garg, “A
Quantitative Survey of various Fingerprint Enhancement techniques”,
International Journal of Computer Applications, Volume 28, pp.24-29, 2011