Performance Enhancement of Minutiae Extraction Using ... · image is transformed from spatial to frequency domain. The FFT based enhancement algorithm includes of the following subsequent

Performance Enhancement of Minutiae Extraction

Using Frequency and Spatial Domain Filters

R.Anandha Jothi Research Scholar

Department of Computer Applications

Alagappa University Karaikudi.

TamilNadu, India.

[email protected] 1

Dr.V.Palanisamy

Professor & Head Department of Computer Applications

Alagappa University Karaikudi.

TamilNadu, India. [email protected]

2

Abstract— Minutiae based feature extraction method is

an important system for person identification. Yet spurious and

false minutiae are often occurring. The fingerprint images are

hardly in good quality. They may be corrupted due to deviation in

skin condition (dry, wet) and skin impression. Therefore,

fingerprint image enrichment techniques are working prior to

minutiae extraction to accomplish more reliable estimation of

minutiae positions. So to overcome these issues, in this proposed

work, spatial and frequency domain filters are effectively

implemented in fingerprint image pre-processing processes

individually and followed by minutiae are extracted. Subsequently

fingerprint image quality is verified in terms of MSE, PSNR and

SSIM and found to be good for AMF. Average value of

performance evaluation in minutiae extraction is found be 0.31and

0.10 for FFT and AMF respectively. Hence FFT can be used

effectively in fingerprint minutiae extraction in person

authentication.

Keywords— Fingerprint; minutiae extraction; Fast Fourier

Transform; spatial domain filter

I. INTRODUCTION

Most of the person identification systems used

biometric characteristics such as behavioral and physical traits

of a person. This supports identification of individual among

groups of others. The fingerprint is one of the most reliable

biometric features. As it remains unique for a specific

individual and has also been verified to be more perfect.

Therefore, fingerprint helps to be the maximum acceptable,

widespread and matured biometric characteristics. A

fingerprint is concerned with pattern of interleaved ridges and

valleys. The group of local characteristic and their relationship

are determining the individuality and invariant features of the

fingerprint. The most often used fingerprint features are

minutiae, Minutiae has two significant characteristics such as

ridge ending and ridge bifurcation that organize a fingerprint

pattern. The fingerprint feature extraction process carried out

by enhancement or pre-processing, minutiae detection and

extraction. Therefore, minutiae based fingerprint

authentication method offers reasonable identification speed

and accuracy. Also, minutiae based feature extraction method

is well acknowledged around the crime and forensic

investigation organizations of maximum countries. The

minutiae map contains around 70 to 100 minutiae points and

matching accuracy is reduced while the size of the database is

growing up. Hence, it is unavoidable to make the size of the

fingerprint feature code to be lesser than possible; finally the

identification may be considerably faster and easier.Automatic

fingerprint identification system (AFIS) for minutiae based

technique usually consists of the following phases (i) Image

pre-processing (ii) Feature detection and extraction. In this

work a comparative study was conducted on the Performance

Evaluation (PE) of extracted minutiae from the two

approaches such as frequency and spatial filters. The rest of

the paper is planned as follows: Section II describes

Frequency domain filter, i.e. fast Fourier Transformation

(FFT) method for fingerprint enhancement. Section III

presents the methods of spatial nonlinear filters (Min/Max,

midpoint, Adaptive Median Filters (AMF)) for fingerprint

enhancement. Section IV The minutiae extraction process

were deliberated. Section V gives experimental results and

discussions are reported. Section V provides conclusion of this

paper.

II. RELATED WORK

Fingerprint image enhancement or pre-processing techniques

are studied by various researchers for removing noise and

other ambiguous effects, Chaohong Wu [1] et.al. Proposed

directional median filters to eliminate noise and spurious

signals for fingerprint image enhancement. Shlomo Greenberg

et al. [2] studied two different methods for removing noise and

improve the quality of fingerprint images. The first method

carried out by wiener filtering for binarization and thinned

image. The second method uses an anisotropic filter for direct

grayscale enhancement. Both methods were improving the

International Journal of Pure and Applied MathematicsVolume 118 No. 7 2018, 647-654ISSN: 1311-8080 (printed version); ISSN: 1314-3395 (on-line version)url: http://www.ijpam.euSpecial Issue ijpam.eu

647

minutiae detection process. Choudhart et al. [3] studied

frequency and spatial filtering for local orientation; frequency

estimation and morphological approach are utilized to improve

the clarity and structure of the ridge structure of the given

input image. B.G.Sherlock et.al. [4, 5] Studied two type filters

for fingerprint image pre-processing. The non-stationary

directional Fourier filter for image enhancement and

Directional filters for image smoothing. The experimental

results of this technique expressively better for AFIS.

E.Chandar et al. [6] tested the median filter on both binary and

gray scale image to make better image quality and the

performance is calculates by statistical correlation and

computational time. This is easiest process but unable to

remove the maximum of noise.

III. ENHANCEMENT METHODS

Fingerprint enhancement is a technique of improving

the quality of an image by increasing brightness, contrast and

sharpening of the minutiae, furthermore to eliminate noise and

other unwanted discontinuities. Accordingly enhancement

process is a major role in pre-processing. Enhanced image

would become try to preserve true minutiae as possible.

Simultaneously destroying the ambiguous patterns such as

spurs break and noise digital image enhancement methods

categorize two types (i) Frequency domain filtering method

(ii) Spatial domain filtering methods. The process of

frequency domain techniques based on modifying Fourier

transform of image [7]. The spatial domain methods referred

to image plane and based on direct manipulation of pixels [6,

8, and 9].The proposed framework as given below fig.1.

Fig. 1 Schematic Diagram of pre-processing

A. Fast Fourier Transform (FFT) for enhancement

Fast Fourier transform is a technique where the given input

image is transformed from spatial to frequency domain. The

FFT based enhancement algorithm includes of the following

subsequent steps (i) Normalization (ii) Segmentation (iii)

Orientation image estimation (iv) 2D Fourier transform (v)

Inverse 2D Fourier transform and reconstruction. The original

image was subjected to contrast enhancement achieved by

Adaptive Histogram Equalization (AHE). This helps in further

essential process and found better minutiae extraction [10].

Additionally the enhanced image was subjected to FFT

filtering, the filtered image were using easier recognition of

ridges and key features [11,12]. The FFT based pre-processed

image was partitioned into overlapping tiny blocks. Normally

the square block size was taken 2k

where k is a whole number.

The power of 2 is taken so that fast radix-2 FFT can be used

and thus optimized the speed. The FFT of the block was

computed by the equation (1,2 and 3).

𝐹 𝑢, 𝑣 =

𝑀−1

𝑥=0

𝑓 𝑥, 𝑦 ∗ 𝑒𝑥𝑝

𝑁−1

𝑦=0

−𝑗2𝜋

× 𝑢𝑥

𝑀+

𝑣𝑦

𝑁 … (1 )

To enhance a particular block by its dominant frequencies,

multiply FFT of the block by its magnitude a set of times as.

𝑔 𝑥, 𝑦 = 𝐹−1 𝐹 𝑢, 𝑣 × 𝐹 𝑢, 𝑣 𝑘 …… (2)

Where 𝐹−1 (𝑓(𝑢, 𝑣)) is given as:

𝑓 𝑥, 𝑦 =1

𝑀𝑁

𝑀−1

𝑢=0

𝐹 𝑢, 𝑣

𝑁−1

𝑣=0

∗ 𝑒𝑥𝑝 𝑗2𝜋 × 𝑢𝑥

𝑀+

𝑣𝑦

𝑁 … . (3)

𝑓𝑜𝑟 𝑥, 𝑦 = 0,1,2… .31

Where K is constant, a higher value of ―k‖ can increase the

presence of the ridges by filling up tiny holes in ridges,

however highest value of ―k‖ may result in spurious ridges

causes an endpoint to become a bifurcation or respectively. An

FFT based fingerprint enhancement or pre-processing is

shown in fig.2.

Fig. 2 (a) AHE image (b) FFT image (c) Binarized image

(b) (a)

(c) (d)

International Journal of Pure and Applied Mathematics Special Issue

648

(d) Extracted Minutiae image

B. Spatial Filtering

Further Spatial domain filtering based enhancement

method was used for pre-processing, the outcome of filtering

produce faithful minutiae [13]. Most of the filters are used for

image noise removal. Which is used for different task i.e.

noise reduction, interpolation, re-sampling and blurring.

Blurring is used to remove small dots and unwanted structures

from an image earlier to large object extraction. The selection

of filter depends upon the type and amount of noise presented

in an image subsequently those dissimilar filters can remove

different types of noise effectively. The spatial domain filters

has two kinds, such as linear and Non-linear filters, in this

work we have achieved by Non-linear Min/Max, Midpoint

filter and Adaptive Median filters (AMF).Non-linear filters

exhibit well performance than linear filters [14].The non-

linear filters functionality is based on the ranking of pixels

enclosed in the image area covered by the filter, and then

replaces the value of middle pixel with the value determined

by the ranking result. The type of spatial domain filters as

follows.

C. Min and Max Filter

The smallest and the largest value within the pixel values

are selected by the minimum and maximum filter. In this

process, the maximum and minimum intensity values are

found within a window of certain pixels. The output remains

unchanged when the central pixel lies within the intensity

range of its neighbors. If the central pixel is greater than the

maximum value within that window, then it is set to the

maximum value and if it is less than the minimum value than

it takes the minimum value to itself [15]. The min/max filter

can be represented by the following equation (4) and (5)

𝑓 𝑥, 𝑦 = max(𝑠,𝑡)∈𝑆𝑥𝑦

{𝑔 𝑠, 𝑡 } … (4)

𝑓 𝑥,𝑦 = min(𝑠,𝑡)∈𝑆𝑥𝑦

{𝑔 𝑠, 𝑡 } … (5)

D. Adaptive Median Filter (AMF)

The AMF is used to determine which pixels in an image

have been pretentious by impulse noise. The AMF categorizes

pixels as noise and then matching each pixel in the given

image to it’s around neighbor pixels. The size of the

neighborhood is modifiable, In addition to the threshold for

the comparison. A pixel that is dissimilar from a majority of

its neighbors, in addition to not structurally align with these

pixels to which it is similar, is called as impulse noise. That

kind of noise pixels is replaced by the median pixel value of

the pixels in the neighborhood that have passed the noise

labeling test.

Notation

𝑍min = 𝑆𝑥𝑦 is the minimum\ lowest grey level value .

𝑍𝑚𝑎𝑥 = 𝑆𝑥𝑦 is the maximum\ highest grey level value.

𝑍𝑥𝑦 = (𝑥,𝑦) grey level at the subject matter

coordinates.

𝑍𝑚𝑒𝑑 = is the maximum possible 𝑆𝑥𝑦window size.

Algorithm

We should analyze the following definition of AMF

at two levels. Let us the function A and B.

Level A: A1 =𝑍𝑚𝑒𝑑 − 𝑍min

A2 =𝑍𝑚𝑒𝑑 − 𝑍max

If A1 > 0 and A2 < 0 then do to definition B.

Level B: B1=𝑍𝑥𝑦 − 𝑍min

B2=𝑍𝑥𝑦 − 𝑍max

If B1 > 0 and B2 < 0, output 𝑍𝑥𝑦 .

Else output 𝑍𝑚𝑒𝑑 .

From the above level A and B of AMF effectively to eliminate

the impulse noise .The definition of Level A and B are given

below.

Explanation of level A: To determine whether the output

value of the filter 𝑍𝑚𝑒𝑑 is in the noise or not If 𝑍min <

𝑍𝑚𝑒𝑑 < 𝑍max the 𝑍𝑚𝑒𝑑 value is not a noise value ant it

must be transmitted to the exit.

Explanation of level B: To determine whether 𝑍𝑥𝑦 𝑖𝑡self is

a noise level and a new value to be according to this. The

definition of the above algorithm that the median value of

level A is equal to the noise, in case similar this the window

size to be inspected will be changed and alternative median

value will be calculated. The above process will be continued

up to the median value comes different between maximum or

minimum value. However, this can’t guarantee that the

obtained value is not noise. Nevertheless, dependent on the

window size the probability of finding a noise value will be

minimized. While increase the widow suppresses the noise to

an excessive extent, simultaneously, proportion to its size.

E. Midpoint Filter

The midpoint filter is used for blurring the image by

replacing it with the average of the highest intensity pixel and

the lowest intensity pixel within the specific window. Median

filter can be signified by the below equation (6).

𝑓 𝑥, 𝑦 =1

2max

(𝑠,𝑡)∈𝑆𝑥𝑦{𝑔 𝑠, 𝑡 + min

(𝑠,𝑡)∈𝑆𝑥𝑦{𝑔 𝑠, 𝑡 } … (6)

F. Minutiae Extraction

Once the frequency and spatial domain filtering has been

applied for enhancing the ridge pattern, then some more steps

are performed for minutiae extraction. Such as Binarization

and Thinning, Binarization of the image obtained is done

keeping the threshold as zero [16, 17]. A value of

representation that the pixel is white and a value of zero

indicate the pixel to be black. A threshold value is set for each

pixel in the image. Those pixel values which are smaller than

International Journal of Pure and Applied Mathematics Special Issue

649

the threshold is set to zero and the pixel values which are

greater than the threshold is set to one. The image produced is

in binary form. Thinning operation is done on the binary

image obtained from the previous step. Thinned ridge lines are

obtained using morphological thinning operator bwmorph [18,

19]. Finally minutiae extraction is done on the thinned image.

This process is done by using a 3x3 window to examine the

local neighborhood of each ridge pixel in the image. The

minutiae extraction is based on crossing number concept.

G. Minutiae Extraction Using Crossing Number (CN)

The crossing number (CN) at a point P is defined as half of the cumulative successive differences [19, 20] and is expressed as follows in equation (7).

CN =1

2 |Pi − Pi+1|

8

𝑖=1

(7)

Where Pi is the value in the neighborhood of P. Pi= (0,1) and I has a period of 8, that is P9=P1.For any pixel P, we consider the 8 neighboring pixels in a 3x3 neighborhood, each of which can take a value of either 1 or 0 as follows: CN has the following properties.

CN Property

0 Isolated Point

1 Ridge Ending Point

2 Continuing Ridge Point

3 Bifurcation Point

4 Crossing Point

The end points and branching points or minutiae are detected using the above mentioned properties of CN. The skeleton image is scanned for detecting the minutiae. It has been found that for ―valid‖ bifurcation points an additional condition of a number of neighbors equal to three is required.

VI. PERFORMANCE EVALUATION (PE)

To evaluate the performance of enhancement filters, the

perceived minutiae are compared with a set of minutiae

attained from the same fingerprint by a human expert. The

performance evaluation (PE) is expressed by the following

equation (8).

𝑃𝐸 =𝑀𝑝−𝑀𝑚 −𝑀𝑠

𝑇𝑜𝑡𝑎𝑙 𝐻 (8)

Wherever Mp is the number of pairs matched within the

enhanced image, Mm and Ms Represent the number of

dropped, spurious minutiae respectively, and TotalH is denoted

as the number of minutiae extracted by the human experts.

The maximum value of PE=1, All the total number of

minutiae are correctly paired with the corresponding detected

minutiae and there are no missing minutiae (Mp=TotalH and

Mm=Ms=0) respectively [21, 22]. Hence all the detected

minutiae are matched or paired. A large value of PE for a

fingerprint image denotes that the enhancement filters have

done a good performance on the image.

A. Quality Measure

After preprocessing the fingerprint image quality was

measured by various quality measurement techniques. Those

techniques are used to estimate the enhancement effect. The

most universally accepted objective measures are i) Structural

Similarity Index Measure (SSIM) ii) Peak signal to noise ratio

(PSNR) and iii) Mean Squared Error (MSE).

B. Structural Similarity Index Measure (SSIM)

The structural similarity index method is based on

measurement of three components such as luminance,

structure and contrast comparison. SSIM process is combined

with these separate components. The structural similarity

amongst the input image x and enhanced image y [23, 24, 25].

SSIM was calculated by the equation (9).

Where 𝜇𝑥 , 𝜇𝑦 are means and𝜎𝑥2, 𝜎𝑦

2 are variances of x and y

respectively, covariance of x, y is 𝜎𝑥 ,𝜎𝑦

and 𝐶1,𝐶2 are

adaptable constants, and L is the utmost possible value of x.

The input fingerprint images from FVC2004 DB3 database

was taken for measuring the SSIM index between original and

enhanced images.

C. Mean Squared Error (MSE)

MSE is calculated by an average squared intensity of the

original and resultant image pixels [26]. There was computed

by the equation (10). Here e (m, n) is error difference between

the original and the distorted images.

𝑀𝑆𝐸 =1

𝑁𝑀 𝑒(𝑚,𝑛) 2

𝑁−1

𝑛=0

(10)

𝑀−1

𝑚=0

D. Peak Signal to Noise Ratio (PSNR)

SNR is a mathematical measure of image quality based on

the pixel difference between input and resultant images. The

SNR to measure the quality of the reconstructed image

compared with the original image. Where S=255 for an 8-bit

pixel image. The PSNR is fundamentally the SNR while all

pixel values are equivalent to the maximum possible value

[27]. PSNR value was calculated by the given equation (11).

𝑃𝑆𝑁𝑅 = 10𝑙𝑜𝑔𝑆2

𝑀𝑆𝐸 (11)

VII. RESULT AND DISCUSSION

A. Quality Metrics

To measure the quality of frequency and spatial filter

based on enhanced fingerprint image. Yet there are many

techniques to evaluate the enhanced fingerprint image quality

such as MSE, PSNR, SSIM that indices depict in table 1 to 4

.The comparison of filters explained by charts shown in fig 3,4

and 5 respectively. The AMF based PSNR and SSIM values

are improved and the value of MSE reduced to compare with

International Journal of Pure and Applied Mathematics Special Issue

650

other filtering method. AMF based enhanced PSNR, SSIM

indices attains the maximum possible values against other

filtering methods. This clearly indicates that the AMF method

enhances gives a better quality fingerprint image as compared

to other filters.

Table 1. Quality measures for min\max filters.

Finger print image Min\Max Filters

DB2 MSE PSNR SSIM

DB3_B_102_7.tif 198.329 22.3152 0.8365

DB3_B_103_5.tif 117.386 24.4153 0.8591

DB3_B_104_7.tif 97.567 26.4695 0.8771

DB3_B_105_2.tif 196.562 22.5421 0.8382

DB3_B_106_1.tif 190.505 22.0283 0.8241

DB3_B_107_3.tif 225.300 23.1250 0.8525

DB3_B_108_5.tif 250.568 20.4862 0.8291

DB3_B_109_7.tif 230.021 21.3549 0.8328

DB3_B_110_8.tif 210.185 20.3110 0.8152

DB3_B_110_9.tif 197.574 21.3841 0.8388

Table 2. Quality measures adaptive median filters.

Finger print image Adaptive Median Filter

DB2 MSE PSNR SSIM

DB3_B_102_7.tif 98.7945 37.1524 0.9897

DB3_B_103_5.tif 94.8033 40.4026 0.9952

DB3_B_104_7.tif 83.1192 40.3852 0.9912

DB3_B_105_2.tif 88.3536 37.2702 0.9857

DB3_B_106_1.tif 100.2391 36.1271 0.9891

DB3_B_107_3.tif 110.3678 35.2491 0.9754

DB3_B_108_5.tif 96.9082 35.0842 0.9711

DB3_B_109_7.tif 90.2107 36.2341 0.9823

DB3_B_110_8.tif 89.1081 33.1782 0.9701

DB3_B_110_9.tif 101.4592 35.0247 0.9714

Table 3. Quality measures for Midpoint

Finger print image Midpoint

DB3 MSE PSNR SSIM

DB3_B_102_7.tif 220.235 21.2215 0.8168

DB3_B_103_5.tif 190.544 21.4356 0.7582

DB3_B_104_7.tif 160.236 22.3548 0.8971

DB3_B_105_2.tif 223.225 19.6544 0.878

DB3_B_106_1.tif 230.754 19.2756 0.802

DB3_B_107_3.tif 252.184 21.4599 0.8576

DB3_B_108_5.tif 263.698 18.415 0.8025

DB3_B_109_7.tif 245.541 17.3542 0.7897

DB3_B_110_8.tif 253.145 19.3633 0.7928

Table 4. Quality measures for FFT

Finger print image FFT

DB3 MSE PSNR SSIM

DB3_B_102_7.tif 128.327 25.3571 0.9728

DB3_B_103_5.tif 115.321 25.5786 0.9835

DB3_B_104_7.tif 121.604 26.6036 0.9615

DB3_B_105_2.tif 160.596 24.5242 0.9677

DB3_B_106_1.tif 135.448 24.8351 0.9782

DB3_B_107_3.tif 112.574 25.5623 0.9736

DB3_B_108_5.tif 134.553 25.1203 0.9685

DB3_B_109_7.tif 130.788 25.4128 0.9745

DB3_B_110_8.tif 157.752 24.5531 0.9352

B. Performance Evaluation

To compare the performance of both filters based on

extracted minutiae. The input image was selected from a

standard fingerprint database FVC2004 of DB2 & DB3. Table

5 & 6 depict the maximum and minimum value of PE of FFT

and AMF for this data set are 0.43- 0.11 and 0.21- 0.03,

similarly with an average value of PE is 0.31 to 0.10

respectively. The PE value calculated more than 100

fingerprint images. Hence, Table 5 & 6 shows only 6 value

those values are randomly chosen based on the number of

spurious minutiae. Such minutiae are grouped by average

(between 13 to 15), below average (between 11 to 12) and

above average (between 16 to 17). The statistical comparison

of those filters, FFT based PE value is much better than AMF

based PE value.

Table 5. PE value for FVC2004 DB2& DB3 database of five

fingerprint images for FFT filtering based extracted minutiae.

Fingerprint image T Mp M

Ms PE d e

DB2_B_101_1.tif 30 28 2 1 17 0.30

DB2_B_106_3.tif 33 28 4 3 11 0.39

DB3_B_110_1.tif 26 25 7 2 15 0.11

DB3_B_101_3.tif 32 29 3 2 12 0.43

DB2_B_109_7.tif 23 22 4 3 13 0.21

DB3_B_105_2.tif 37 35 2 3 17 0.43

Average PE value 0.31

Table 6. PE value for FVC2004 DB2 & DB3 database of five

fingerprint images for adaptive median filtering (AMF) based

extracted minutiae.

Fingerprint image T Mp M

Ms PE d e

DB2_B_101_1.tif 30 25 5 1 18 0.06

DB2_B_106_3.tif 33 26 6 3 13 0.21

DB3_B_110_1.tif 26 22 4 2 16 0.07

International Journal of Pure and Applied Mathematics Special Issue

651

Fig. 3 MSE value for

Spatial and frequency domain filters

Fig. 4 PSNR value for

Spatial and frequency domain filters

Fig. 5 SSIM value for

Spatial and frequency domain filters

DB2_B_110_1.tif

26 21 3 3 17 0.03

DB3_B_101_3.tif 32 25 7 2 15 0.09

DB3_B_105_2.tif 37 31 6 3 19 0.16

Average PE value 0.10

VIII. CONCLUSION

Fingerprint image enhancement has been successfully

studied using frequency and spatial domain filtering

techniques. Herein, AHE, Filtering, Binarization, Thinning

and minutiae extraction applied effectively on individual

fingerprint and implemented in MATLAB. The quality of the

reconstructed images is determined by measuring the MSE,

PSNR, and SSIM and performance is evaluated using DB3,

DB2 of FVC2004 database. The proposed fingerprint

enhancement system using spatial filtering techniques gives

high PSNR, SSIM when compared to the frequency filtering.

However, frequency domain filter based PE value is higher

when compared to the spatial filtering method.

References [1] Chaohong Wu, Sergey Tulyakov and Venu Govindaraju, Image

Enhancement Method using Directional Median Filter, in Proc.

SPIE conf. on Biometric Technology for Human Identification.

[2] Greenberg-―Fingerprint Image Enhancement Using Filtering

Techniques‖-Pattern Recognition, 2000, proceedings 15th

International Conference, Volume 3, page(s)322-325 . [3] J.Choudhary, Dr.S.Sharma, J.S.Verma, ―A new framework for

improving low quality fingerprint images,‖ international journal of

computer technology and application. Vol.2, no.6, pp.1859 -1866,2011.

[4] Schneider, J.K., Glenn,W.E., 1996. Surface Feature Mapping

Using High Resolution C-Span Ultrasonography. US Patent 5587533.

[5] Sherlock, D., Monro, D.M., Millard, K., 1994. Fingerprint

enhancement by directional Fourier filtering. In: IEEE Proc. on Visual Imaging Signal Processing, vol. 141, pp. 87–94.

[6] E.Chandra et.al Noise Suppression Scheme using Median Filter in

Gray and Binary Images, International Journal of Computer Applications (0975 -8887) Volume 26–No.1, July 2011

[7] Rafael C. Gonzalez, Rechard E. Woods, Digital Image Procesing,

Pearson,Third Edition,2008. [8] E.Chandra and K.Kanagalakshmi, Noise Elimination in Fingerprint

Images using Median Filter, International Journal of Advanced

Networking and Applications,(2011),Vol 02, Issue:06, pp:950-955. [9] K.Kanagalakshmi and E.Chandra, Performance Evaluation of

Filters in Noise Removal of Fingerprint Image, Proceedings of

ICECT-2011, 3rd International Conference on Electronics and Computer Technology, April 8-10 2011, pp. vol.1:117-123, ISBN:

978-1-4244-8677-9, Published by IEEE, Catalog no.: CFP1195F-

PRT, IEEE Explore. [10] Youlian Zhu, Cheng Huang, An Adaptive Histogram Equalization

Algorithm on the Image Gray Level Mapping. Gray Level

Mapping ―, 2012 International Conference on Solid State Devices and Materials Science, Physics Procedia 25 (2012) 601 – 608

1875-3892 © 2012 Published by Elsevier. Pp. 601 – 608.

[11] Neethu S, Sreelakshmi S, Deepa Sankar ―Enhancement of fingerprint using FFT×|FFT|n filter‖ International Conference on

International Journal of Pure and Applied Mathematics Special Issue

652

Information and Communication Technologies (ICICT 2014) 2015

The Authors. Published by Elsevier , ( 2015 ) 1561 – 1568 [12] Abdullah Toprak a, ˙Inan Guler , Impulse noise reduction in

medical images with the use of switch mode fuzzy adaptive

median filter, Digital signal processing, Digital Signal Processing 17 (2007) 711–723.

[13] Divya.V-―Adaptive Fingerprint Image Enhancement Based On

Spatial Contextual Filtering And Preprocessing Of Data‖-IJCAT journal Volume1,Issue 4,2014

[14] K. Somasundaram and P. Kalavathi, ―Medical Image Denoising

using Non-Linear Spatial Mean Filters for Edge Detection.‖, rural.univ.ac.in, pp. 149-153

[15] R.C.Gonzalez and R.E.Woods, ―Digital Image Processing,‖ second

ed., Prentice Hall, Englewood, Englewood, Cliffs, NJ, 2002. [16] Ratha, N.K., Chen, S., Jain, A.K.: Adaptive flow orientation-based

feature extraction in fingerprint images. Pattern Recognition.

28(11), 1657–1672 (1995). [17] Zhixin Shi, Venu Govindaraju: A chaincode based scheme for

fingerprint feature extraction.Pattern Recognition Letters. 27, 462–

468 (2006). [18] Tico, M., Kuosmanen, P.: An algorithm for fingerprint image post-

processing. In: Proc. of the Thirty-Fourth Asilomar Conference on

Signals Systems and Computers, pp. 1735-1739 (2000). [19] Zaho, F., Tang, X.: Preprocessing and post-processing for

skeleton-based fingerprint minutiae Extraction. Pattern

Recognition. 40(4), 1270–1281 (2007). [20] Ching-Tang Hsieh, Eugene Lai, You-Chuang Wang, An effective

algorithm for fingerprint image enhancement based on wavelet transforms Pattern Recognition 36 (2003) 303 – 312

[21] D. Peralta, M. Galar, I. Triguero, O. Miguel-Hurtado, J.M. Benitez,

F. Herrera, Minutiae filtering to improve both efficacy and efficiency of fingerprint matching algorithms, Elsevier,

Engineering Applications of Artificial Intelligence 32(2014) 37-53.

[22] Alan C. Brooks, Xiaonan Zhao, Thrasyvoulos N. Pappas, Structural similarity quality metrics in a codingcontext: exploring

the space of realistic distortions, IEEE Transactions on image

processing, vol. 17, no. 8, Aug. 2008 1 [23] D. Maltoni, D. Maio, A.K. Jain and S. Prabhakar, Handbook of

Fingerprint Recognition, Springer, New York, 2009

[24] Z. Wang and A. C. Bovik, ―A universal image quality index,‖ IEEE Signal Process. Lett., vol. 9, no. 3, pp. 81– 84, Mar. 2002.

[25] Z. Wang and A. C. Bovik, Modern Image Quality Assessment. San

Rafael, CA: Morgan and Claypool, 2006. [26] WU Xiao-Feng, HU Shi-gang, ZHAO Jin, LI Zhi-ming, LI Jin,

TANG Zhi-jun, XI Zai-fang, Comparative analysis of different

methods for image enhancement, © Central South University Press and Springer-Verlag Berlin Heidelberg 2014 , J. Cent. South Univ.

(2014) 21: 4563−4570, DOI: 10.1007/s11771-014-2461-y

[27] L. M. Jean-Bernard Martens, "Image dissimilarity," Signal Processing, vol. 70, no. 3, pp. 155-176, 1998.

International Journal of Pure and Applied Mathematics Special Issue

653

654