A05610109

IOSR Journal of Engineering (IOSRJEN) www.iosrjen.org
ISSN (e): 2250-3021, ISSN (p): 2278-8719
Vol. 05, Issue 06 (June. 2015), ||V1|| PP 01-09
International organization of Scientific Research 1 | P a g e
Finger Knuckle Analysis: Gabor Vs DTCWT
Hanna Paulose, Er. Harbinder Singh
M.Tech Student Department of ECE SUS College of Engineering and Technology Tangori, Mohali, India
Assistant Professor Department of ECE SUS College of Engineering and Technology Tangori, Mohali, Punjab
Abstract: - Knuckle biometrics is one of the current trends in biometric human identification which offers a
reliable solution for verification. This paper analysis FKP recognition based on the behaviour of two different
filtering and classification methods. Firstly,Gabor Filter Banks techniques are applied for finger knuckle print
recognition and then the same database is analysed against Dual Tree Complex Wavelet Transform technique.
The experiment is evaluated to identify finger knuckle images using PolyU FKP database of 7920 images.
Finally, these two different systems are comparedfor false acceptance rate FAR,true acceptance, false rejection
rate FRR and true rejection. Extensive experiments are performed to evaluate both the techniques, and
experimental results show the pros and cons of using both the techniques for specific applications.
Keywords: - Finger Knuckle Print (FKP); Gabor Filter; Dual Tree Complex Wavelet Transform;
I. INTRODUCTION
Use of biometrics dates back over a thousand years, which is defined as the measure of human body
characteristics such as fingerprint, voice pattern, Iris and Hand measurement, Gait, Palm Print. It is a potent and
unique tool based on the physiological (anatomic) and behavioralcharacteristics of the human beingsas shown in
Fig.1.
Figure 1: Types of Biometric System
Typically, we can identify the following attributes of a biometric system such as
Uniqueness:It means that an identical feature should not appear in two different people.
Universality:This is the feature which ispresent/occurs in as many people as possible ie universally.
Permanence: This feature does not change over time,or we can say that it contrast very slowly.
Measurability: The possibility to measure the feature with relatively simple technical instruments
Acceptability:Its is the feature of the measurement in daily lives by all.
Circumvention:It is the toughness of the system to be deceived by fraudulent methods.
This paper discusses about the biometric identifier finger knuckle print (FKP), which is the outeror
exteriorsurface of the fingers of a person ie the phalangeal joint. Thisconfiguration or pattern of the exterior of
knuckle phalangeal joint is unique and dissimilar and can be used for authentication very efficiently and
effectively. Pattern of the finger knuckle is highly discriminative and makes the surface a unique biometric
identifier. The Finger Knuckle print recognition system typically contains data acquisition, ROI extraction,
feature extraction, coding and matching process. For the comparison universally accessiblePolyU FKP database
of 7920 images were utilized to study the features using different techniques and algorithms. The features of FKP

are initiallyextracted using the Gabor Filter Banks and then the same database images are independently filtered
using Dual Tree Complex Wavelet Transform (DT-CWT).
II. RELATED STUDIES
The local characteristics or information of FKP was accessed using Scale Invariant Feature Transform
SIFT and Speed up Robust Features SURF technique. The SIFT Scale Invariant Feature Transform algorithm is
used to get the important ie key points using the scaling and invariant features which were matched to prove the
user authentication [1]. Ajay et.al, studied and suggested method of triangulation for authentication using hand
vein images of all finger’s finger knuckle s[2].Tao Kong et al.[3]proposed a extraction and classification method
based on hierarchical model for finger knuckle print recognition, rooted froman establishedstylefusion method of
scores obtained from the images. In the method which is under analysis, firstly Gabor feature has been taken as
the basic feature for finger knuckle print recognition and then a new decision rule is defined based on the
predefined threshold.G S Badrinath et al.[1] proposed a system which uses feature extraction with the help of
features that are local of improved FKP that are sifted out using the scale invariant feature transform and the
speeded up robust features extraction techniques. Nearest neighbour ratio technique is used for matching of
features of finger knuckle print and then matching scores of SIFT and SURF are fused using the weighted sum
rule.Lin et al. [4] using the outcomes of psycho-physics and neuro-physiology studies that show that image
perception is based on both local and global features, suggested, the Local Global Information Combination ie
(LGIC) technique. For local feature extraction and classification, the orientation information from the message is
extractedusingfour scales and six orientations Gabor filters and is encoded using the modest coding technique.
This process is appropriate for images containing ample line like constructions and has the benefits like
greataccurateness, sturdiness tobrightnessdisparities, and easy andfast matching. Subsequently the scale of the
Gabor filter is amplified to infinity by which the Fourier transform of the finger knuckle print image is obtained.
The Fourier constants of the image are taken as the global feature. For matching two good,competitivecode maps,
normalized Hamming distance obtained from used angular distance is used. BLPOC ieBand Limited Phase Only
Correlation is used to measure the correspondence between Fourier transforms ie Global Information of the
images. Thus both features ie the local and global features are extracted and the matched separately and two
distances obtained, d1 and d2 are then fused according to the rule distance given in Matcher Weighting (MW)
algorithm.
III. FKP RECOGNITION SYSTEM ARCHITECTURE
Figure 2: Schematic Diagram for FKP recognition
The system consists of FKP image acquisitionassembly followed by data processor whichconsists of
Region of Interest extraction, featureextraction and coding and finally matching isperformed to differentiate data
into imposter orgenuine. The study comprises of comparing two different filtering techniques.
Figure 3: (a,b) finger knuckle print images. (c,d) Region of Interest images.

IV. FEATURE EXTRACTION AND MATCHING
This study features study of two different extraction technique and comparison between both of them.
The technique of the dual tree complex wavelet transform is initiated by the reason that it provisionsin removing
the effects of non-uniform brilliance, and the directional features provided by the dissimilar sub bands makes it
possible to identify edge information with different directionalities in the equivalent image. The consequential
complex-waveletbased feature information are more discriminating and distinguishing compared to the Gabor
wavelet-resultant features and at the same time are of lower dimension and time consuming when compared with
that of Gabor wavelets. 2-D dual-tree complex wavelet transform is more effective, less time complex and
computationallycompetent. Furthermore, in addition to the reliable and capable classification performances,
method 2-D dual-tree complex wavelet transform has a really low computational complexity.
Methodology
The study uses 2D Gabor filters and 2D Dual Tree Complex Wavelet Transform (DT-CWT) to extract
the image local orientation information. This information is then employed to extract and represent the finger
knuckle print recognition features. While matching of two knuckles, the Euclidian distance or hamming distance
is used to measure the similarity.
Feature Extraction
Feature extraction is very important step in pattern recognition because of poor quality and its necessary
to capture all options in finger knuckle image for associate economic and competitive and reasonable matching.
The effective approachesfor matching two coarsetexturedlike biometric pictures is to decide their identical or
similarity in spectral domain depiction of mistreatment the section information. Firstly, intra-class
differencesshould be small i.e. features derived from different samples of the same class should be close enough.
Secondly, the inter-class separation or differences should be large which means that features derived from
samples of different classes must differ significantly to ensure there is no matching between two different
knuckles. Also information or feature should be independent of the size, rotation and location of the pattern. 2D
Dual tree complex waveletis independent with regards to translation, rotation and scale, and therefore can be used
for pre-processing or extracting featuresof images.
Gabor Filtering Feature Extraction
The Gabor filtering technique can extract the spatial-frequency and orientation information from the
original signal [5]. It has been extensively used as a convolution filter for feature extraction to fulfill the feature
extraction job. For years together Gabor Filter technique has been broadly used as an effective toolfor feature and
information extraction job in face recognition, iris verification, fingerprintanalysis and palm print recognition
systems. In [6], Loris and Alessandra described a Gabor feature selection technique which can produce features
that are wavelength, phase, magnitude, orientation and bandwidth, which can be used separately or jointly in
different combinations in different systems. Kong calculated the three basic three features of Gabor Filter Banks
(magnitude, phase, and orientation) produced by the Gabor filter for face recognition system and concludes that
the orientation feature is the most robust, discriminative and distinctive feature [7]. T.S. Lee [8] evaluated the
situations under which a set of continuous 2D Gabor wavelets will provide a complete illustration of any image,
and also determinesimilar to the self image basedwavelet features and parameterizations which allow constant
reconstruction of the image from the filtered and superimposed images by summation even though the wavelets
formed in an orthonormal basis. In this paper, the 8 filters banks of Gabor filter is proposed by combining the
orientation, wavelength, phase and magnitude features using Gabor filter for knuckle print recognition.
Experiment and analysisresultscompare and verify the scheme against Dual Tree Complex Wavelet Transform
(DT-CWT) based on performance in FKP recognition. The Gabor filter banks has different methods in the
literature and here we adopt the one of the form proposed by Lee [8] as given below:
g 𝑥, 𝑦; 𝑓, 𝜃 = 𝑒
−
1
2
𝑥 𝜃
2
𝜎 𝑥
+
𝑦 𝜃
2
𝜎 𝑦
.cos (2𝜋𝑓 𝑥 𝜃 )
Where
𝑥 𝜃 = 𝑥 sin 𝜃 + 𝑦 cos 𝜃
𝑦 𝜃 = 𝑥 cos 𝜃 − 𝑦 sin 𝜃
and, g 𝑥, 𝑦; 𝑓, 𝜃 denotes the Gabor filter value in spatial domain x & y denote the x and y coordinates of the
Gabor filters, f denotes the frequency of the sinusoidal plane wave in the direction Ө from x-axis.𝜎𝑥 and 𝜎𝑦 denote
the standard deviations of the Gaussian envelope along x & y-axesseparately. Four parameters (θ, f,𝜎𝑥 , 𝜎𝑦)of
Gabor filtermust be specified for applying it to an image. The frequency of the filter is completely determined by
the local ridge frequency and the orientation is determined by the local ridge orientation. The selection of the

values 𝜎𝑥 and 𝜎𝑦 involves trade off. The lager the values of 𝜎𝑥 and 𝜎𝑦 , the more stable and robust the filters are to
the discrepancies and noise in the fingerprint clip/ image, but also are more likely to generate spurious undesired
ridges and valleys.
On the other hand, the smaller the values of the filter, the less likely the filters are to introduce spurious
ridges and valleys but then they will be less effective in removing and clearing the noise. If𝜎𝑥 and 𝜎𝑦 are standard
deviations of the Gaussian envelope and the values are too large, the filter is more stable and robust to noise, but
is probable to smoothens the image to the extent that the ridges and valley details in the fingerprint are lost. If
𝜎𝑥 and 𝜎𝑦 readings are very small or less, the filter is not operative and effective in removing the noise. Gabor
technique is one of the common choice for texture and pattern analysis. There are different kinds of popular sets
of Gabor filters. These sets are generally designed based on illustration considerations. The 2D filter represented
by extracts the orientation and magnitude information at each pixel in digital knuckle image.At given threshold,
the probability of accepting imposter as correct person known as false acceptance rate FAR and likelihood of
rejecting genuine user as imposter is known as false rejection rate FRR are calculated. Equal Error Rate EER is an
ideal rate or optimal rate where FAR is equal to false rejection rate i.e FRR. Diagrammatically, Equal Error Rate,
EER is known or characterized as the crossing point or meeting point between false acceptance rate FAR and
false rejection rate FRR. It is commonly used to determine the overall accurateness of the system and serve as
comparative measure against the other biometric systems.
Dualtree Complex Wavelet Transform
Complex wavelet filters in two dimensions offer twodirectionalselectivity. So they are able to segregate
and differentiate all part of the two dimensional frequency space being implemented separately as two distinct
systems. Complex coefficient at each level aligned at angles of ±15º,±45º,±75º are generated from six bandpass
images. These features and characteristics are useful for pattern recognition of finger knuckle print. In this study
we present a new finger knuckle feature extraction method by using 2D DTCWT. The feature extraction scheme
is to use the multi-level coefficient of decomposition part of normalized finger image via 2D DT-CWT, which is
implemented using dual tree complex wavelet structure. There are many different types of techniques suggested
in the literature for extracting unique and invariant feature on rotation, translation and scale. Wavelet techniques
are used in the wide range of problems in extraction, classification, data compression, de-noisingand
reconstruction. Some procedures and techniques used at the output of the wavelet transform to generate a binary
feature vector, while other technique consider real value feature vector as output as a real valued featured vector.
But the ordinary discrete transform is not shift invariant because of the decimal operation during the transform.
Therefore any minor shift in the input signal can cause different output coefficients. DT-CWT has improved
directionality and reduced shift sensitivity and it is approximately invariant. It has real and imaginary part in the
coefficients. The DT-CWT has two real wavelet transform in parallel where the wavelets of one branch are the
other are Hilbert transform of the wavelet. In any high security application, both extremely low falseacceptance
rate FAR and false rejection rate FRR are desirableat the same time this is also called the doublelow problem.
The discrete wavelet transform (DWT) which is commonly used in its maximallydecimated form. This technique
works good for compression of images.Its use for analysis of signal and reconstruction of image tasks has been
hindered by two disadvantages of discrete wavelet transform DWT:
 DWT has lack of shift invariancewhich means that a small shift in the incoming signal can causea significant
variation in the distribution of energy of coefficients of DWT at different scales.
 For the diagonal features it has poor directional and rotational selectivity, as the filters arereal and
distinguishable.
Shift invariance can be provided by use of the undecimated form ofthe dyadic filter tree compared to the
decimated discrete wavelet transform DWT tree.However, this suffers from increased computationrequirements
compared to the fully decimated discrete wavelet transform DWT and also exhibits high redundancyin the output
characteristics/ features, making successive processing costlyalso.
Dual-tree complex wavelet transform (DT-CWT) has less time complexity as it is more computationally
efficient.When DT-CWT is applied to multi-dimensional signals it has a good directional
sensitivity.Summarizing the properties of DT-CWT:
 Approximate shift invariance;
 Good directional selectivity in 2-dimension (2-D) like Gabor filter banks.
 Perfect reconstruction of the image (PR) using short linearphase filters;
 Computational Efficiency.
The complex wavelet methods, analysed here, offer perfect reconstruction,better directional selectivity, and a
general multiscale decomposition.

V. RESULTS & DISCUSSIONS
The database for FKP images has been provided by Biometrics Research Center, The Hong Kong
Polytechnic University. The database consists of165 FKP images having images of four fingers and 12 images of
each finger at different ambient lights. Experiments were carried out to appraise the functioning, evaluate and
compare the performanceof both methods. Experimental results verifythat the DT-CWT method can achieve
satisfactoryrecognition in real time. For two techniques, differentfeature extraction result in different
recognitionperformances. Finally, these two systems are compared for the features extraction viz the False
Rejection Rate, False Accept Rate and Equal Error Rates. Experiments are performed to evaluate both the
techniques and capabilities, and experimental results demonstrate that DT-CWT can achieve anencouraging
performance.
Figure 4: Performance of Gabor extraction technique
Figure 5: Performance of DTCWT extraction technique
Comparison Gabor Filter vs Dual Tree Complex Wavelet Transform
DT CWT is less time complex and computationally efficient system produced the following outputs and
allows the transform to provide approximate shiftinvariance anddirectionally selective filters while preserving the
properties of perfect reconstruction and computationalefficiency with good well-balanced frequency responses.
Further the degree of threshold and feature detection capability is more with the Dual tree complex wavelet
transform.
False Reject vs True Accept
The False Reject Vs True Accept graph shows that for a low value of the false reject the standards of true accept
is better for DTCWT technique. This means that the DTCWT system performance is better compared to Gabor
technique.
-10
0
10
20
30
40
50
60
0 10 20 30
Percentage
Threshold
Gabor Filter Performance
False Reject
False Accept
True Accept
True Reject
-20
0
20
40
60
80
0 500 1000
Percentage
Threshold
DTCWT Perfomance
False Reject
False Accept
True Accept
True Reject

True Reject vs True Accept
The True Reject Vs True Accept graph shows that for a low value of the false reject, the readings of true
accept is better for DTCWT technique. This means that the DTCWT system performance is better compared to
Gabor technique.
False Accept vs False Reject
The False Accept Vs False Reject graph shows that the false accept of DTCWT is lower than the Gabor System.
Thisanalysis depicts that the DTCWT system performance is better compared to Gabor technique.
0
20
40
60
80
-20 0 20 40 60
TrueAccept
False Reject
Gabor vs DTCWT
Gabor
DTCWT
0
20
40
60
80
-20 0 20 40 60
TrueAccept
True Reject
Gabor vs DTCWT
Gabor
DTCWT
-10
0
10
20
30
40
50
60
-20 0 20 40 60
FalseReject
False Accept
Comparison: Gabor vs DTCWT
Gabor False
Reject
DTCWT False
Reject

False Accept vs True Reject
The False Accept Vs True Reject graph shows that the false accept of DTCWT is lower than the Gabor System.
This illustrates that the DTCWT system performance is better compared to Gabor technique.
False Accept vs True Accept
The False Accept Vs True Accept graph shows that the false accept of DTCWT is lower than the Gabor
System and also better True Accept values are obtained for a low FalseAccept in case of DTCWT technique .
This means that the DTCWT system performance is better compared to Gabor technique.
False Accept vs All Values of Gabor and DTCWT
The False Accept Vs other values in the graph shows that the false accept of DTCWT is lower than the Gabor
System and also better True Accept values are obtained for a low False Accept in case of DTCWT technique .
This means that the DTCWT system performance is better compared to Gabor technique.
-10
0
10
20
30
40
50
60
-50 0 50 100
TrueReject
False Accept
Gabor True
Reject
DTCWT True
Reject
0
20
40
60
80
-20 0 20 40 60
TrueAccept
False Accept
Gabor True
Accept
DTCWT True
Accept
-20
0
20
40
60
80
-50 0 50 100
Percentage
False Accept
Gabor False
Reject
Gabor True
Reject

0 2 4 6 8 10 12 14 16 18 20
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
Time Complexity of Gabor and DTCWT
Time Complexity of Gabor and DTCWT in the graph shows that the time required to compute DTCWT features
is 10 times lesser than the Gabor System and also time elapsed to compute is for DTCWT technique is stable
barring the time taken to compute for initial two subjects. This means that the DTCWT system performance is
better compared toGabor.
VI. CONCLUSION AND FUTURE WORK
This paper presented a comparison of two approaches for personal identification using two dimensional
2D finger knuckle image. Feature based approach 2D Gabor filter method and 2DDual tree complex wavelet
transformis used for identification of knuckle based system.2D DT-CWT bases systems for feature extraction and
classification have been quite effective and efficient in achieving high performance in knuckle recognition and
classification. Further, the Dual tree complex wavelet transform Finger Knuckle Identification technique has
several advantages such as less time complexity, computational efficiency, user friendliness, reasonable size and
cost effectiveness. It has a great potential to be improved and employed in real time applications or utilities in
future. This work deliberated single finger knuckle image for identification and classification. In future this can
also be extended by considering multiple finger knuckle images along with bending characteristics. Finger
geometry and knuckle bending information may also be combined and used for person identification. Variations
in the knuckle creases due to some disease or disorder may degrade the performance, which requires further
investigation.
VII. ACKNOWLEDGMENT
I extend my heartfelt thanks to my husband, Jaimandeep Singh, parents in law and my son Prathyush whose
love, patience, and continual support helped me to complete this study.
REFERENCES
[1] G.S Badrinath, Aditya Nigam and Phalguni Gupta, “An Efficient Finger-knuckle-print based Recognition
System Fusing SIFT and SURF Matching Scores” , Proceedings of 13th International Conference on
Information and Communications Security.
[2] Ajay Kumar and Yingbo Zhou, “Personal Identification using Finger Knuckle Orientation Features”,
Electronics Letters Vol. 45, No. 20.
0 2 4 6 8 10 12 14 16 18 20
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Time Complexity: DTCWT
time
time
Time Complexity: Gabor
Time Complexity: DTCWT

[3] Tao Kong, Gongping Yang and Lu Yang, “A hierarchical classification method for finger knuckle print
recognition”, EURASIP Journal on Advances in Signal Processing 2014 http://asp.eurasipjournals.com/
content/2014/1/44
[4] LinlinShen, Li Bai and Zhen Ji, "Hand-based biometrics fusing palmprint and finger knuckle-print"
[5] D. Gabor, Theory of communication, Journal of the Institute of Electrical Engineers 1946; 429-457.
[6] L. Nanni, A. Lumini, On selecting Gabor features for biometric authentication, International Journal of
Computer Applications in Technology.
[7] A. Kong, “An evaluation of Gabor orientation as a feature for face recognition”, Proceedings of ICPR’08.
[8] T.S. Lee, “Image representation using 2D Gabor wavelet” IEEE Trans. Pattern Analysis and Machine
Intelligence1996.

A05610109

More Related Content

What's hot (20)

Viewers also liked (7)

Similar to A05610109 (20)

More from IOSR-JEN (20)

Recently uploaded (20)

A05610109