International Journal of Engineering and Techniques - Volume 3 Issue 5, Sep - Oct 2017

RESEARCH ARTICLE OPEN ACCESS

A Novel Finger Knuckle Print Recognition Algorithm Using Radon

Coefficients
1
ChandrakiranSahu, 2Prof.YogeshRathore
1,
M.Tech Research Scholar,Computer Science Department,RITEE,Raipur
2,
Asst. Prof,Computer Science Department, RITEE,Raipur

Abstract:
Person authentication is one of the main concerns in industries as well as in the academics because it is used in these
area for various application such as Computer security, banking enforcement, law enforcement physical access control etc.
Biometrics is basically referred as the unique behavioral and psychological characteristics of human which can be used for
distinguishing the individuals and hence can be served as one of the optimum solution to this problem.In this thesis, inner
knuckle surface and its pattern are used for developing a system for person identification. A database of the finger knuckle
containing 7920 samples collected from 660 individuals that is publicly made available by the Hong Kong Polytechnic
University is used in this work. Radon transform is used here to extract the features of the inner knuckle surface as it is most
widely used for detection of lines and crease. A database of these extracted feature is prepared and various techniques of
classification is used for classifying these pattern and hence identification of person. Comparison has also been made between
various classification technique to find out the most accurate identification.

Keywords — Biometrics, finger knuckle print, features, recognition system.

I. INTRODUCTION canuniquely define each member of large

population which makes them suitable for
Biometric recognition or simply biometric large scale identification. Reasonof attraction
refers to use of distinctive anatomical and of such traits is social acceptance and easy to
behavioral characteristicsfor automatically use. Finger knuckle (FK) is user centric,
recognizing an individual. Automatic human contactless andunrestricted access control. Its
identification has become an important issue texture and statistical features are available
intoday’s global information society. Due to and easily extracted. It is independent toany
increasing security concerns, large number of behavioral aspect[3-8].
systems currently requiredpositive
identification before allowing an individual to Woodward and Flynn are the first scholars
use their services. During the last decade there who made use of the finger knuckle surface in
has been asteady research effort toward their work. They set up a 3D finger back
providing user friendly and reliable surface database with the Minolta 900/910
methodologies for access to facilities, research sensor.
andservices. Automatic biometric systems
have emerged as a more reliable alternative to C.Ravikanth et al.[19] developed a system for
the traditional personalidentification acquiring the finger back surface images. This
systems[1][2]. Different techniques have been imaging system uses a digital camera focused
developed each of them having its own against a white background under uniform
advantages anddisadvantages Now a day’s illumination.
authentication is based on the unique Lin Zhang et al. [1] development a system for
physiological and behavioral characteristics of FKP acquisition. This consists of four
human being.It is generally accepted that components FKP image acquisition, ROI
physical traits like iris, fingerprints, finger (region of interest) extraction, feature
knuckle, finger vein, DNA finger print extraction and feature matching.

ISSN: 2395-1303 http://www.ijetjournal.org Page 116

 International Journal of Engineering and Techniques - Volume 3 Issue 5, Sep - Oct 2017

Fusion is a promising technique that is used to

Some of the noteworthy contribution can be increase the accuracy of the biometric systems
found in the paper [2, 3, 14, 27] that can [2,4,6,8,20,27].
create spatially localized features are
receiving increasing attention in the literature. II. METHODOLOGY
Jun et al.[14] proposed a new linear feature As mentioned earlier that Radon transform
extraction approach called Weighted Linear can be used for detecting the unique line and
Embedding(WLE). crease in the finger knuckle print image even
for noisy image therefore in this project work,
Yang et al. [2] inspired by the work that radon transform is used to extract the features
Gabor wavelets have been applied i.e. Radon transform coefficients
icients which are
successfully in image analysis and pattern unique for unique lines and crease. Though
recognition, used it for feature representation hough transform can also be used in place of
in FKP. Jing et al.[3]simultaneously the radon transform because it is also able to
considered distances and angles between detect the lines and crease in the image but in
image data vectors to measure data similarities
simil case of noisy image its performance is poor.
in hope of more sufficiently capturing the
manifold structure. Different coding based A. RADONTRANSFORM
algorithms arealso proposed in the literature. In recent years the Hough transform and the
[5, 13, 15, 17] and basically iris code is the related Radon transform have received much
foundation of these coding algorithms. These attention. These two transforms are able to
coding techniques have been used widely for transform two dimensional images with lines
palmprint recognition [21, 22, 23, 24, 25, 29] into a domain of possible line parameters,
and have provided good recognition results. where each line in the image will give a peak
Ajay Kumar [17] in his work exploited the positioned at the corresponding line
local information in comparison to the global parameters. This have lead to many line
information for reliable performance. Lin et processing, computer vision, and seismic.
al.[13] designed
signed a system to capture FKP
detection applications within image
images and proposed a method to align the
FKP images by adaptively constructing a local
coordinate system for each image. Lin et al.
[15] proposed a fast feature extraction and
coding method called the Monogenic code
based on the Monogenic signal theory and is
used for FKP recognition. Lin Zhang et al. [5]
used coding method because they have the
merits of high accuracy, robustness, Figure 1 Line parameter using Radon
compactness and high matching speed. Hence, Transform
based on the findings that Riesz transform can
well characterize the visual patterns this work Several definitions of the Radon transform
proposes to encode the local patches of FKP exists, but the are related, and a very popular
images by using second order Riesz form expresses lines in the form
transform. Some other noteworthy other
methods can be found in the literature [7, 8,
9,11, 16] various image processing techniques B. CREATION OF FEATURE DATABASE
aree employed either independently or Image acquired by the camera now a days are
combined to extract the texture, local, global mostly colour camera and Radon transform
or line feature from the finger knuckle print.
ISSN: 2395-1303 http://www.ijetjournal.org Page 117
 International Journal of Engineering and Techniques - Volume 3 Issue 5, Sep - Oct 2017

can be used in gray scale image only as it is Step 1 Input the colour Finger Knuckle print
defined for only two dimensional signal. image.
Therefore it is imperative to convert the Step 2 Convert the image from colour to gray
acquired colour image in to the gray scale scale image by using colour to gray scale
image. The process of converting the colour operation.
image in to a gray scale image is called RGB Step 3 Resize the image for getting the same
to gray scale conversion.When image is number of radon coefficients.
acquired during the acquisition process then Step 4 Apply Histogram operation to Make
there is the possibility of acquired image is the contrast elevation.
contrast-wise not uniform which again pose Step 5 Apply the Radon transform and
the problem of poor recognition accuracy compute the Coefficients for the step of three
therefore it is necessary to make the contrast degree angle. And store the
of the image uniform, this work is coefficients corresponding to the each angle in
accomplished by the histogram equalization the database.
operation. Number of radon transform Step 6 Store the Coefficients of each image in
coefficients are same for the image of the the feature database.
same size and different for the different size of Step 7 End of operation.Once all the features
the image. Different number of coefficients are extracted then the next step is to classify
i.e. features creates the different dimensional all the features and hence recognize the person
size database which pose a problem of on the basis of the classification. Though there
matching therefore in this project work image are so many classification method, but in this
is first of all resize to some standard work we have concentrated on 4 different
dimension size so that we can get the same classification methods. Out of four
number of radon transform coefficients. classification methods two are neural network
Following are the Algorithm Steps of feature based and two are the pure classifier. Block
Database Creation diagram of the classification is shown in the
figure 4. Following are the steps of
classification and identification

Step 1 Input the colour Finger Knuckle print

image.
Step 2 Convert the image from colour to gray
scale image by using colour to gray scale
operation.
Step 3 Resize the image for getting the same
number of radon coefficients.
Step 4 Apply Histogram operation to Make
the contrast elevation.
Step 5 Apply the Radon transform and
compute the Coefficients for the step of three
degree angle.
Step 6 Perform the classification Task using
BPN neural network and compute the
accuracy.
Step 7 Perform the classification task using
RBFN neural network and compute the
accuracy.
Figure 2 Block Diagram of Database Creation
Step 8 Perform the classification task using
ISSN: 2395-1303 http://www.ijetjournal.org Page 118
 International Journal of Engineering and Techniques - Volume 3 Issue 5, Sep - Oct 2017

the KNN classifier and compute the knuckle print, 10 features are extracted.
accuracy.Step 9 Apply the naïve bias Feature database id prepared. Back
classifier for the classification and compute propagation neural network is designed for
the accuracy. classification these radon coefficients.Neural
Step 10 Compare the performance of the network is then trained. Supervised learning is
above mentioned classifier with the base paper used to train the neural network. Since in
classifier and draw the conclusion supervised learning, target vector is used
therefore target vector is also designed for the
training. Feature vector is given to the neural
network as an input and target vector is given
to the network as the output. Once the neural
network is trained it is hen tested for different
finger knuckle print.test image.

Figure 4 GUI Designed for the FKP

recognition System

Once the feature are extracted then

classification is performed using BPN neural
network, RBFN neural network, KNN
classifier, and Naive Bias classifier for the
classification and compute the accuracy.
Figure 3Block Diagram of classification &
Recognition Task Table 1Ten- different Radon coefficient for
different value of Theta

III. EXPERIMENTAL RESULTS

As mentioned earlier that Radon transform
can be used for detecting the unique line and
crease in the finger knuckle print image even
for noisy image therefore in this project work,
radon transform is used to extract the features
i.e. Radon transform coefficients which are
For performance comparison, accuracy is
unique for unique lines and crease. 10 features
computed which is defined as
of the finger knuckle are extracted out from
the finger knuckle. Each of these 10 feature
actually represent the average radon
coefficient taken for angles which are 18
degree apart. This created total 10
feature(180/18). Therefore for each finger It very important to see that how the different

ISSN: 2395-1303 http://www.ijetjournal.org Page 119

 International Journal of Engineering and Techniques - Volume 3 Issue 5, Sep - Oct 2017

classification method perform for the similar Figure 6 Accuracy Comparison for different
number of features therefore apart from the
BPN network, three more classification ClassifierFrom the above graph it is clear that
methods were also tested i.e. RBFN RBFN network perform better than rest of the
network(Radial basis Function Network) , classification method. Table 2compare the
KNN (k-nearest neighbourhood ), and Naïve performance of the best classifier in the
bias method for the similar number of the proposed work to the best performer
feature and the accuracy obtained is tabulated classification method of the paper[31].|
in the table 2
Table 3 Comparison of method with the
Method No. of Accuracy
Method Proposed in the paper[31]
Feature (in %)
Paper 10 94.33
[31]Implementation Accuracy
Method with RBFN 10 97.25
network in this 98
97
work 96
95
94
93 Accuracy
Table 2 Accuracy Table for different classifier 92
Paper[31] Method Presented
S. No. Classification Accuracy Implementation in this paper
No. Features Method (in %)
1 10 BPN 94.5
Figure 5.12 Comparisons Graph of
4 10 RBFN 97.25 Accuracy Between paper[31] and method
7 10 KNN 93.25 implementation
10 10 Naïve Bias 91.25
III CONCLUSION
From the table 2it is clear that RBFN is found
to be the best classifier with accuracy of the Finger knuckle print is comparatively new
97.25%. rest of the classifier are found to be biometric characteristics of the human being
lesser accurate in recognizing the finger which is being explored by many researcher
knuckle print and hence the person. for recognition and authentication purpose.
Radon transform is one of the important tool
98 which can be used to detect the lines pattern in
97 any image. This property of the Radon
96 transform is used in this project for extracting
Accuracy

the features of the finger knuckle print image

95
BPN and then utilized it for person recognition.
94
RBFN Accuracy of about 97.25 is obtained by using
93 this methodology which is exceptionally well
92 KNN
and proved it as a very good tool for the
91 Naïve Bias recognition of a person. This work also
90 present the performance comparison of
various classifier i.e. BPN, RBFN, KNN and
Number of
Naïve Bias.From the result obtained it is
Features=10
evident that RBFN is best in term of accuracy
of recognition among all the above mentioned
ISSN: 2395-1303 http://www.ijetjournal.org Page 120
 International Journal of Engineering and Techniques - Volume 3 Issue 5, Sep - Oct 2017

classifier and therefore is most suitable for on Single Knuckle print Image”, Asia-Pacific
Conference on Information Processing, pp. 218-
classification purpose. 221,2009.
[16] Ajay Kumar and Yingbo Zhou, “Personal
Identification using Finger Knuckle Orientation
REFERENCES Features”, Electronic Letters Vol.45, no 20, pp.1-
[1] Lin Zhang, Lei Zhang, David Zhang and Hailong Zhu,
7,September 2009.
“Online Finger-Knuckle –Print Verification for
[17] Zhang Lin “Personal authentication using finger
Personal Authentication” Pattern Recognition, vol. 43,
knuckle print” doctoral diss., The Hong Kong
pp. 2560-2571, 2010
Polytechnic University. Jan 2011
[2] YANG Wanknou, SUN Changyin and SUN Zhongxi,
[18] D.L. Woodard, P.J. Flynn, Finger surface as a
“Finger-Knuckle-Print Recognition Using Gabor
biometric identifier, Computer Vision and Image
Feature and OLDA”, Proceedings of the 30th Chinese
Understanding 100 (3) (2005) 357– 384.
Control Conference, pp.2975-2978,July 22-24, 2011.
[19] A. Kumar, C. Ravikanth, Personal authentication using
[3 ] Xiaoyuan Jing, Wenquian Li, Chao Lan, Yongfang,
finger knuckle surface, IEEE Trans. Information
Yao, Xi Cheng and Lu Han, “Orthogonal Complex
Forensics and Security 4 (1) (2009) 98-109.
Locality Preserving Projections based on Image Space
[20] A. Ross, A.K. Jain, Information fusion in Biometrics,
Metric for Finger-Knuckle-Print Recognition”, 2011.
Pattern Recognition Letters 24 (13) (2003) 2115–2125
[4 ] AbdallahMeraoumia, SalimChitroub and Ahmed
[21] W.K. Kong, D. Zhang, Palmprint texture analysis
Bouridane, “Fusion of Finger-Knuckle-Print and
based on low-resolution images for personal
Palmprint for an Efficient Multi-biometric System
authentication, in: Proceedings of 16th International
of Person Recognition”, Proceedings of ICC, 2011.
Conference on Pattern Recognition, vol. 3, 2002, pp.
[5 ] Lin Zhang, Hongyu Li and Ying Shen, “A
807–810.
NovvelReisz Transforms based Coding Scheme for
[22] A.W.K. Kong, D. Zhang, Competitive coding scheme
Finger-Knuckle-Print Recognition”, Proceedings of
for palmprint verification, in: Proceedings of
International Conference on Hand Based
International Conference on Pattern Recognition, vol.
Biometrics,2011.
1, 2004, pp. 520–523.
[6] Yanqiang Zhang, Dongmei Sun and ZhengdingQiu,
[23] A. Kong, D. Zhang, M. Kamel, A study of brute-force
“Hand-based single sample biometrics recognition”,
break-ins of a palmprint verification system,
Proceedings of ICIC, 2010.
IEEESystems, Man and Cybernetics, Part B 36 (5)
[ 7] ZaharaS.Sharaiatmadar, KarimFaez, “A Novel
(2006) 1201–1205.
Approach for Finger-Knuckle-Print Recognition Based
[24] A. Kong, K.H. Cheung, D. Zhang, M. Kamel, J. You,
on Gabor Feature Fusion”, 4th international Congress
An analysis of Biohashing and its variants,
on Image and Signal Processing, pp.1480-1484, 2011.
PatternRecognition 39(2006) 1359–1368.
[8 ] ZHU Le-quing, “Finger Knuckle print recognition
[25] D. Zhang, W.K. Kong, J. You, M. Wong, On-line
based on SURF algorithm”, Eight International
palmprint identification, IEEE Transactions on Pattern
Conference on Fuzzy Systems and Knowledge
Analysis and Machine Intelligence 25 (9) (2003) 1041–
Discovery (FSKD), pp.1879-1883, 2011.
1050
[ 9] GohKahOng Michael and Tee Connie and Andrew
[26] A.Rattani, D.R. Kisku ,M.Bicego, and M. Tistarelli
TeohBeng Jin, “ Robust Palm Print and Knuckle
feature level fusion of face and fingerprint biometrics .
Print Recognition System Using a Contactless
Inbiometrics: theory, Applications and systems pages
Approach”, 5th IEEE Conference on Industrial
1-6, 2007.
Electronics and Applications, pp.323-329, 2010.
[27] ShubhangiNeware, Kamal Mehta, A.S. Zadgaonkar,
[10] Lin Zhang, Lei Zhang, Davind Zhang and Hailong Zhu,
Finger Knuckle Identification using
“Ensemle of local global information for finger-
PrincipalComponentAnalysisandNearest Mean
knuckle-print recognition”, Pattern recognition,vol. 44,
Classifier, International Journal of Computer
pp. 1990-1998, 2011
Applications, Volume 70- No.9, May 2013
[11] LinlinShen, Li Bai and Zhen Ji, “Hand-based
[28] Choras M, Kazil R, “Knuckle Biometrics Based on
biometrics fusing palmprint and finger knuckle-print”,
Texture Features”, International Workshop
[12] Lin Zhang, Lei Zhong and David Zhang, “Finger-
onChallenges for Hand-Based Biometrics (ETCHB),
Knuckle-print: A New Biometric Identifier”, ICIP
IEEE, 2010.
2009, pp1981-1984, 2009.
[29] Kumar A and Zhou Y, “Human identification using
[13] Jun Yin, JingboZhong Jin and Jian Yang,
knuckle codes”, Proceedings BTAS, Washington,
“Weighted Linear Embedding and Its Application
2009.
to Finger Knuckle-Print and Palmprint
[30] Damon L. Woodard, “Exploiting finger surface as a
Recognition” “Proceedings of the International
biometric identifier”, doctoral diss., Notre Dame,
Workshop on Emerging Techniques and Challenges
Indiana December 2004
for Hand Based Biometrics, 2010.
[31] Dixit Vaibhav, Patil M. Pradeep,
[14] Lin Zhang, Lei Zhang and David Zhang, DeshpandeArvind“Person Recognition Based on
“Monogenic Code: A Novel Fast Feature Coding Knuckle Print Biometric Features Computed using
Algorithm with Applications to Finger-Knuckle- Radon Transform” IJARCSSE , Volume 4, Issue
Print Recognition “Proceedings of the International 3,March 2014
Workshop on Emerging Techniques and Challenges
for Hand Based Biometrics, 2010.
[15] Rui Zhao, Kunlun Li, Ming Liu and Xue Sun, “A
Novel Approach of Personal Identification Based

ISSN: 2395-1303 http://www.ijetjournal.org Page 121