狮子座天秤座SIViP
DOI10.1007/s11760-013-0436-3
ORIGINAL PAPER
Combining Gaborfilter and FFT forfingerprint enhancement bad on a regional adaption method and automatic gmentation Morteza Zahedi·Ozra Rostami Ghadi
Received:6April2012/Revid:15November2012/Accepted:1February2013
©Springer-Verlag London2013
Abstract Fingerprints are the best biometric identity mark due to the consistency during life time and uniqueness.To increa the classification accuracy offingerprint images,it is necessary to improve image quality which is a key role for correct recognition.In other words,enhancing thefingerprint images leads us to obtain better results in classification offin-gerprint images.Although Gaborfilter and fast Fourier trans-form(FFT)are ud to enhancefingerprint images,Gabor filter acts better than FFT in detection of incorrect ridge endings and ridge bifurcation,while FFT tries to connect broken ridges together andfill the created holes.This paper tries to enhance gray-scalefingerprint images by
combin-ing the Gaborfilter and FFT in order to get benefit from the advantages of each enhancingfilter(Gaborfilter and FFT).
A method is propod forfingerprint image gmentation bad on the image histogram and density.By employing the propod method which enhances thefingerprint images using the better enhancingfilter in each part,the experimen-tal results show that the wholefinger print is better enhanced, and conquently,it leads to a better recognition rate. Keywords Fingerprint enhancement·Gaborfilter·FFT·Image quality·Regional adaptation·
Automatic gmentation
M.Zahedi(B)·O.R.Ghadi
School of Computer Engineering and Information Technology, Shahrood University of Technology,Shahrood,Iran
e-mail:zahedi@shahroodut.ac.ir
O.R.Ghadi
e-mail:Rostami_
1Introduction
Afingerprint is made of ridges and valleys.The ridges are the dark area in afingerprint,while the white area existing between the ridges are called valleys(e Fig.1a,(Refs.[4, 22])).
Fingerprint classification depends strongly on the quality of the inputfingerprint images(Ref.[1]).Actually,a notable percentage of recordedfingerprint images available for pub-lic are the images with poor quality.In poor-qualityfinger-print images,the ridge structures are not always well-marked, and therefore,they cannot be properly detected for recogni-tion(Refs.[6,8]).This caus to appear following problems:•A considerable number of counterfeit minutiae may be formed,
•A significant percent of primary minutiae may be disre-garded,
•High mistakes in minutiae localization may appear.
An example offingerprint image of very poor quality is shown in Fig.1b that ridge structures are completely destroyed.The aim of an enhancement algorithm is improv-ing the clarification of ridge structures of inputfingerprint images,and sofingerprint images more accurately be clas-sified(Refs.[5,9]).Fingerprint enhancement can be imple-mented on following groups:
1.Binaryfingerprint images,
2.Gray-levelfingerprint images.
In a gray-levelfingerprint images,ridges and valleys in a local neighborhood construct a sine-formed wave which has a well-specified frequency and orientation.
One of the most extensive techniques ud forfingerprint enhancement is the method employed by Hong et al.(e Ref.[3]),which is bad on the complexity of the image with
SIViP
幼师自我介绍Fig.1a Fingerprint image,b fingerprint images of very poor quality
Gabor filters adapted to the local ridge orientation and ridge frequency.The Hong’s method is bad on the guesstimated local ridge orientation and frequency which improves the clarification of ridge and valley structures in input fingerprint images.
Gabor filters have two important properties of frequency-lective and orientation-lective and conquently cau an optimal common resolve in both spatial and frequency domains.So,using of Gabor filter as a band-pass filter makes it suitable to delete the noi and maintain true ridge/valley structures (Refs.[7,15]).The technique of fast Fourier trans-form filter is ud for the enhancement of fingerprint images which aims at improving the fingerprint image by conjoin-ing some faultily broken points on ridges and removing some spurious connections between ridges.
Although Gabor filter and fast Fourier transform (FFT)are ud to enhance fingerprint images,Gabor filter acts better than FFT in detection of incorrect ridge endings and ridge bifurcation,while FFT tries to connect broken ridges together and fill the created holes.This paper tries to enhance gray-scale fingerprint images by combining the Gabor filter and FFT in order to get benefit from the advantages of each enhancing filter (Gabor filter and FFT).
In the following ctions,the propod method for the enhancement of fingerprint images is described in details.Section 2address the main steps of the algorithm.Sec-tion 3describes two basic methods of fingerprint enhance-
ment (Gabor filter and fast Fourier transform).The propod method and results of the implemented fingerprint enhance-ment algorithm are shown in Sect.4where the experimen-tal results are performed on FVC2004data t containing 322fingerprint images.The propod method for fingerprint image gmentation bad on image histogram and density can be expresd in Sect.5.Finally,Sect.6contains the sum-mary,conclusion and future work.2Fingerprint image enhancement
A gray-level fingerprint image is determined by an N ×N matrix,where I (i ,j )indicates the intensity of the pixel at the i th row and j th column.The Eqs.(1)and (2)reprent the mean and variance of a gray-level fingerprint image,respec-tively.
Mean (I )=1N
2
xuanwuN −1 i =0N −1
j =0
I (i ,j )(1)
Variance (I )=1N
N −1 i =0N −1
j =0
(I (i ,j )−Mean (I ))2
(2)
An orientation image displays the local ridge orientation at each pixel.Local ridge orientation is usually determined for a block rather than at every pixel;an image is divided into a t of w ×w nonoverlapping blocks and for each block is determined a single local ridge orientation.A frequency image that shows the local ridge frequency is reprented as the frequency of the ridge and valley structures in a local neighborhood along a direction normal to the local ridge orientation.The region mask reprents the category of the pixel.
The algorithm enhanced the ridges and valleys of finger-print images in part of images that is damaged,and this makes the ridge more visible and recognition of ridge endings and ridge bifurcation from each other.The better image enhance-ment algorithm helps appear genuine minutiae and remove the spurious minutia that may be causing errors in matching and classification.
The flowchart of the fingerprint enhancement algorithm is shown in Fig.2.The main steps of the algorithm are described in the next ction.2.1Normalization
An input fingerprint image is normalized such that it has a mean and variance that are already defined.Normalization is a process bad on pixel-wi and does not change the clarity of the ridge and valley structures.The main goal of normalization is to diminish the variations in gray-level val-ues along ridges and valleys (Refs.[10,11]).
SIViP
Fig.2Aflowchart of thefingerprint enhancement algorithm
2.2Local orientation estimation
The orientation image is guesstimated from the normalized inputfingerprint image.The orientation image express an inherent property of thefingerprint images and in a local neighborhood determines invariable coordinates for ridges and valleys of inputfingerprint image.
2.3Local frequency estimation
Normalized inputfingerprint image and the computed orientation image are ud to compute frequency image (Ref.[14]).In a local neighborhood where there are no minu-tiae and singular points,the gray levels along ridges and val-leys can be modeled as a sine-formed wave along a normal direction to the local ridge orientation(e Fig.3(Ref.[3])). So,one of thefingerprint image properties is local ridge.If there are no minutiae and singular points in the oriented win-dow,the x-signature forms a discrete sine-form wave with the same frequency.So,the frequency of ridges and valleys can be computed from the x-signature.
2.4Region mask estimation
羽毛球基本动作
The region mask is created by classifying each block in the normalized inputfingerprint image into a block.A block (or a pixel)in an inputfingerprint image could be either in a recoverable region
or an unrecoverable region.Classification of pixels into recoverable and unrecoverable class can be done bad on the evaluation of the form of the wave created by the local ridges and valleys offingerprint image.Fig.3Oriented window and x-signature
2.5Filtering
Frequency and orientation for structures of parallel ridges and valleys in afingerprint image provide beneficial infor-mation,which is ud in removing undesirable noi.The sinusoidal-shaped waves of ridges and valleys vary slowly in a local constant orientation.So,we can satisfactorily remove the und
esirable noi and maintain the correct ridge and val-ley structures using a band-passfilter that is adapted to the corresponding frequency and orientation.The most famous fingerprint enhancement technique usfilter methods that make ridges obviously differentiated from each another,con-nect broken ridges,fill holes and delete noi.Gaborfilter is suitable to enhanced areas offingerprint images where ridges and valleys are shown pale and of low-density,but FFTfilter is suitable in areas offingerprint images in which ridges are wider,more staying clo together.
格式合同
2.6Binarization
Binarization is the process that changes a gray-level image into a binary image.This betters the contrast between the ridges and valleys in afingerprint image and thus simplifies the minutiae extraction.
Most extraction of minutiae algorithms operate on binary images(Ref,[2]).In binary images,there are only two levels of interest:The black pixels and the white pixels express ridges and valleys,respectively(Ref.[19]).
The overall issue of image binarization has been exten-sively studied in differentfields,such as image processing and pattern recognition.The binarization process checks the each pixel’s gray-level value
in the enhanced image,and if the value is greater than the global threshold,then the pixel value is t to a binary value one;if not,it is t to zero.The outcome
SIViP
is a binary image with two levels of information,the ridges in foreground and the valleys in background (Ref.[13]).
3Filtering methods 3.1Gabor filter
A bank of Gabor filters is adapted to local ridge orientation
and ridge frequency that ud the ridge pixels and valley pixels in the normalized input fingerprint image to create an enhanced fingerprint image (Refs.[17,18]).
The form of parallel ridges and valleys with well-defined frequency and orientation in a fingerprint image contains u-ful information,which is ud in removing undesirable noi.The sine-formed waves of ridges and valleys vary slowly in a local stable orientation.A band-pass filter is adapted to the corresponding frequency and orientation;therefore,it can delete unfavorable noi and maintain the correct ridge and valley structures (Ref.[20]).
Gabor filters have two important properties of frequency-lective and orientation-lective,and it has optimal com-mon resolve in both spatial and frequency domains.So,using Gabor filters as band-pass filters,it is suitable to delete the noi and maintain true ridge/valley structures.General form of Gabor filter is as follows:
H (i ,j :ϕ,f )=exp
−
12 i 2ϕ
δ2i +j 2ϕδ2j cos (2πf i ϕ)(3)i ϕ=i cos ϕ+j sin ϕ(4)j ϕ=i sin ϕ+j cos ϕ
(5)
That ϕis the orientation of the Gabor filter,f is the frequency of a sine-wave,and δi is the space stable of the Gaussian cover along i axis,and δj is the space stable of the Gaussian cover along j axis (Refs.[12,21]).
Clearly,the frequency characteristic of the filter (f )is fully specified by the local ridge frequency,and th小天鹅e10
e orienta-tion characteristic is specified by the local ridge orientation.Thus,the more robust the filter is to the noi,the greater the amount is lected,but there is more probability that the filters make counterfeit ridges and valleys.On the contrary,there is less probability that the filters make counterfeit ridges and valleys,if we lect the smaller values of δi and δj ;in result,they will be less efficient in deleting the noi.The values of δi and δj can be lected according to experimental data.
3.2Fast Fourier transform (FFT)filter
Technique of fast Fourier transform filter is ud for the enhancement of fingerprint image,and general form of fast Fourier transform is equal to:
F (u ,v)=
M −1 i =0N −1 j =0
f (i ,j )×exp
−k 2π×
ui
M +v j N
u =31,v =0,1 (31)
(6)
That size of blocks is u ×v (32×32).
FFT of the block is multiplied by its value a t of times to enhance a specific block by its most frequencies in which the magnitude of the original is FFT =|(F (u ,v))|.Result of the enhanced block is equal to d (i ,j )=F
−1
F
(u ,v)×|F (u ,v)|
h
(7)
That F −1(F (u ,v))is obtained with the following equation:
F (i ,j )=1M N M −1 i =0N −1
j =0
f (u ,v)×exp −k 2π× ui M +
v j
斗车N
i =0,31,j =0,31.(8)
The h is a constant specified by experiment in Eq.(7).If “h ”
is the highest value,then the form of the ridges improves and small holes are filled in ridges (Ref.[16]).
FFT improves to conjoin some faultily broken points on ridges and to remove some spurious connections between ridges,and conquently,the enhanced image will improve.
3.3Combining Gabor filter and FFT
As shown in Fig.4,the images reprent results of enhance-ment fingerprint images using Gabor filter and FFT methods.Fingerprint areas are parated with green and red circles.Green circles demonstrate area of images in which FFT method enhanced better than Gabor filter method,and red cir-cles demonstrate area of images in which Gabor filter method enhanced images better than FFT method.
It ems that the number of incorrect diagnosis of ridge endings and ridge bifurcation is very least for Gabor filter than FFT method.So,Gabor filter detects incorrect ridge endings and ridge bifurcation better than FFT.The num-bers of spurious ridge endings detected and ridge bifurcation detected in Gabor filter method are less.Resolution between the ridges and valley in images that enhanced using Gabor filter is better.Recoverability of ridges with poor quality in Gabor filter is more than FFT method.Using of Gabor filter method is not suitable for areas with high complexity.Gabor filter method is poor in maintaining ridges with much curva-ture,while in this respect,FFT is better and good.Although FFT maintains some fal connections between the ridges,it tries to connect broken ridges together and fill the created holes.
SIViP
Fig.4a,d,g Input images with poor quality;b,e,h enhanced images by FFTfilter method;c,f,i enhanced images by Gaborfil-ter method.Green circles demonstrate area of images in which FFT method enhanced better than Gaborfilter method.Red circles demon-strate area of images in which Gaborfilter method enhanced images better than FFT method(colorfigure online)
Conquently,it can be said that using of Gaborfilter is suitable to enhanced areas offingerprint images where ridges and valleys are shown pale and of low-density,such asfin-gerprint with dry skin;namely,in thefingerprint areas in which ridges are not specified properly,using Gabor gives good results.In areas offingerprint images in which ridges are wider,the more clo they are together,with high curva-ture and much complexity,such as oilyfingerprint,it is better to u FFT method instead of Gaborfilter method.
Thus,according to the expresd contents,in this paper,it is propod thatfingerprint images be divided into gments and regions.Then the gments can be reviewed accord-ing to the characteristics of the FFTfilter and Gaborfilter to lect an appropriate enhancement method for each g-ment of thefingerprint.Then each gment enhanced with appropriate method(FFTfilter or Gaborfilter).In fact,each region is adapted with one of the appropriate methods
and Fig.5a Original image;b enhanced image by FFTfilter method;c enhanced image by Gaborfilter method;d originalfingerprint image is divided.e Enhanced image using propod method
the region offingerprint enhances using it.Propod method works with regional adaptation offingerprint,and each region offingerprint enhances by an appropriate method.
As shown in Fig.5a–c,some parts offingerprint image with high curvature and much complexity enhance using FFT method better than Gabor method.Thus,the originalfinger-print image is gmented into nine gments,and then for gments in which FFTfilter is working better than Gaborfil-ter,FFTfilter is ud to enhance image and for the other parts using Gaborfilter.Figures5d,e show combination of Gabor filter and FFT to enhancefingerprint image.The obtained results are compared with the results of Fig.5a–c,and it shows that the propod method has better performance in fingerprint enhancement.
4Experimental results
The purpo of afingerprint enhancement algorithm is to improve the clarity of ridges and valleys of inputfinger print images and make them more suitable for the minutiae extrac-tion algorithm and the classification algorithm.Usually,bina-rization offingerprint image improves minutiae extraction, and it is ud forfingerprint classification.
七一表彰The goal of the experimental result ction is to display the results of enhancement algorithm.Fingerprint enhancement algorithm is tested on FVC2004databa containing322fin-gerprint images(typical poorfingerprint images).Here,some of the input images and their enhanced images are illustrated; you can e the results as binary images.
