You are on page 1of 4

2018 International Conference on Soft-computing and Network Security (ICSNS)

Evaluation Of Human Age With FKP Using K-NN


A.KaviPriya A.Muthukumar
Department of Electronics and Communication Engineering Department of Electronics and Communication Engineering
Kalasalingam Academy of Research and Education Kalasalingam Academy of Research and Education
Krishnankoil, India Krishnankoil, India
Kavidivya222@gmail.com muthuece.eng@gmail.com

Abstract—Humans have distinctive and unique traits which It has U-shaped joint describing the number of lines and
can be used to distinguish as a form of identification. Biometrics length &spacing between lines. Knuckle has crease pattern
identify people by measuring some aspect of individual anatomy &stray marks on the basis of Photographic identification and
or physiology such as face, fingerprint, iris etc., In this present these features are unique and opt for personal authentication.
work age of several persons are trained from FKP impression It has three joints or bones in each finger named as proximal
and results are analyzed with the performance examination phalanx, PIP and DIP joint. In present era, voting system has
based on the matching ratio in the testing phase which include been considered as high level of acclaimed for under-aged
classification of age from K-NN using voronoi representation voter registration which results in election irregularities. So,
Age estimation is considered as an important investigative
Keywords—Gabor filter; Finger Knuckle Print(FKP); Artificial
Neural Network(ANN), K-Nearest Neighbor algorithm (KNN); Age
leads for finding unknown person. Existing method of age
& Gender Estimation estimation have set mainly for the crime investigation on the
availability of teeth, bones etc.So in this present work, FKP
I. INTRODUCTION has been used as a important biometric trait for age
identification due to its unique nature which does not change
Biometrics technology incorporates the identification
over the life of a soul. Several works shows that age
&authentication of individuals by dissecting the human body estimation is significant factor in identification of individual,
characteristics like Palmprint, Fingerprint, Iris etc. Compared Likewise Hu-Han [1] estimates the Biological inspired
with fingerprint, the finger knuckle surface has some benefits features [BIF] extraction from the real time face datasets and
as a biometric identifier recently, it has been found that the SVM is essential for the extraction of demographic attributes
template pattern of skin folds and crimps in the outer finger that is being highlighted. Gnansivam.P [2] shows that Finger
knuckle surface which is highly rare and thus can avail as a print is taken into account for the estimation of person’s age
peculiar biometric identifier. Foremost it is not easy to be gender with SVD&DWT along with KNN classifier. Ali-
blemished since people usually retain stuffs with the inner side Abdul [3]delivers the result of identification of iris with
of the hand. In addition with various use of fingerprint, there is extraction of Gabor filters on the basis of Average Absolute
no slur of criminal analysis associated with the FKP, so it can Deviation to obtain the matching scores based on genuine or
impostor. Imed Bouchrika [4] explore the histogram of
have a greater user acceptance. Thus, the finger knuckle
features based on the classification of tree-based fashion
feature has a splendid potential to be widely accepted as a which include feature selection process by the basis of
biometric identifier. The matching of finger knuckle patterns discriminative traits for the better cluster efficiency. A.Deepa,
will help to mark the suspects and find out authorizing T.Sasi praba [5]defines the estimation of age by geometrical
scientific proof from the images, particularly in cases once no measures sake of case of median filter for isolating the noise
evidence about fingerprints or face is boon within the market and edge details. Konstantinous [6]denotes the one-
images. This Fig.1 basically has high textured region and it dimensional Gabor filter with centre frequency as Gaussian
has no stigma associated with criminal investigation, voter window of uncertainty principle. Jian Hou [8] utilized multiple
registration etc. kernel learning method in case of optimization process and
usage of KNN with average combination method for better
efficiency and accuracy Rest of the paper is organized as
follows Section 2 introduces the overview of proposed work
and it describes extraction of FKP with Gabor feature and
Section 3 gives the details regarding the estimation of age
using KNN algorithm that includes voronoi representation of
nearest neighbors Experimental results are basically explained
in Section 4 and in Section 5 fully compile the basic work that
provide ideas for future research

Fig. 1. FKP Textured Region

978-1-5386-4552-9/18/$31.00 ©2018 IEEE


2018 International Conference on Soft-computing and Network Security (ICSNS)
II. OVERVIEW OF THE PROPOSED WORK basically use in classification &other related problems like
This work is mainly about the performance of K-NN predictive power, easy interpretation of output, with low
algorithm with FKP image which can be widely used for computation time [7]
various banking application, crime investigation, voter
registration etc. This process involves the matching of FKP to
show the estimated data about age in several forensic
applications that helps mainly for voter registration in
calculating the under-aged persons in case of misbehavior
voters

Fig. 3. Proposed block diagram for computing distance with


KNN

This fig.3 shows the training record of age of unknown


persons.This K-NN algorithm helps in choosing the k value of
FKP inputs for computing the distance based on the euclidean
distance and predict the output based on the k-value for
Fig. 2. Overview of the proposed work in age &gender estimating the age based on voronoi representation for
estimation
calculating the random element with nodes and edge vertices
of FKP.
This fig.2 shows the estimation of age groups in these
two phases,namely training phase and testing phase.Both
features includes the extraction of features using Gabor filter
bank from the database by showing the real and magnitude
parts of FKP.Former training phase includes features which
are further classified according to the computing euclidean
distance based on a K-NN algorithm. Latter in testing phase
the FKP image that is chosen to authenticate matches and
delivers the age of the person with accuracy.

A. Gabor filter extraction of FKP images


Gabor filter is mainly used for Denoising of biometrics,
which helps to remove the corrupted features by changing the
pixel value. Mainly it helps to obtain the localized frequency Fig. 4. Voronoi nodes&vertices representation
information, which is generally determined from low
frequency information. Further tuning of Gabor filter leads to This Fig.4 shows the voronoi representation of FKP for
specific frequency direction obtained based on the number of calculating the euclidean distance to find the nearest neighbor.
degrees. They are well suited in extracting the texture
information of images. Likewise this work also includes the ALGORITHM FOR IMPLEMENTING KNN
extraction of Gabor coefficients to show the real and Training phase
magnitude parts. Train ← KNN (C , D )
 − 1  x '2 y'2  
G ( x, y : f , θ ) = exp  2 + 2   cos(2Πfx') (1) D' ← Pr e Pr ocess(D )
 2 δ x ' δ y '  
K ← Selectk (C , D')
Where
x' = x sin θ + y cos θ (2) return − D' , k
y ' = x cos θ − y sin θ (3) Testing phase
This phase includes the computation of nearest neighbor
And ‘f’ represents the frequency of the sinusoidal plane wave representation in KNN algorithm in estimating age
along the direction of θ from x-axis and δ xδ y is space Apply ← KNN (C , D ' , k , d )
constant of the Gaussian envelope along x, y axes SK ← ( D ' , k , d )
III. ESTIMATION OF AGE USING KNN ALGORITHM foreach ← e j ∈ C
A. K-NN In Estimation Of Age sk ∩ c j
do − p j ←
K-nearest neighbor algorithm is mainly used over non- k
linear problems due to high accuracy output. This can return ← arg max p j

978-1-5386-4552-9/18/$31.00 ©2018 IEEE


2018 International Conference on Soft-computing and Network Security (ICSNS)
This algorithm helps to calculate the nearest points with k
chosen nodes with training and testing that includes the
process of pre-processing of FKP filter to classify the input in
order to estimating age of the person which helps in
classification of objects based on closest training samples.
Considering the classification stage for relating the similarity
function includes Fig. 6. Gabor Filter Bank With Real And Magnitude Parts

This Fig.6 explains the real and magnitude parts of Gabor


(x1 − x2 )
+ ( y1 − y2 )
2 2
D= (4) filter bank in FKP image that produces features which helps in
computing the distance in K-NN
X fkp − Min fkp
N '= (5)
Max fkp − Min fkp
N ' fkp = 1 − N fkp (6)

This helps in calculating the nearest distance formation in


case of similarity function like otherwise denoted as "Lazy
Learning Algorithm" that finds its closest training sample in
case of sample 1 as FKP and the normalization technique Fig. 7. Age estimation in FKP
mainly deals with the matching scores that is generated as
common scores on the basis of similarity index which is This Fig.7 shows the estimated age for each person using
calculated on weight age sum rule. FKP by feature extraction using Gabor filter

< x i , y j > for predicting y j (7) This Table.1 shows the estimated age-groups for various
index and middle fingers that represent the age between 20-60

x − xi
2
(
=  xi − y j )
i 2
(8)
in various dataset
j TABLE.1 AGE ESTIMATION OF FKP

Query instance are classified according to the nearest Age groups in various FKP
j
sample in predicting the y Right index Right middle Left index Left middle

IV. RESULTS AND DISCUSSION 55 55 47 50


35 45 25 60
This execution of the proposed personal age estimation 25 55 20 20
system is evaluated by using two different publicly available 45 35 45 25
biometric databases viz., PolyU finger knuckle print database
41 41 25 41
(PolyU Finger Knuckle Print Database) These dataset consists 50 45 30 43
of finger knuckle print images that are captured using ignoble 20 50 35 45
cost and low resolution camera in a peg free environment. In
28 20 40 20
this database, the FKP images of left index finger, left middle 25 47 41 25
finger, right index finger and right middle finger of each 47 22 50 35
person were acquired. PolyU database consists of totally 7920
images with 660 different finger knuckle images collected The accuracy and Success rate in case of matchng ratio in
from 165 persons. For this age evaluation analysis, the FKP terms of Error are calculated basis of false rate as 0.01,0.1&1
images obtained from 50 subjects are considered as training are calculated for various range of FKP from (54.66 -73.21)
samples and images obtained from 25 subjects are considered in right index ,similarily other FKP are estimated as (59.11-
as test images. 81.13) in left middle,right middle,left middle
TABLE.2 GAR&FAR FOR VARIOUS FKP

ERROR SUCCESS RATE


Right Right Left
Range Left Index
Index Middle Middle
Fig. 5. Experimental Output For Enhancing The FKP Image 0.01 55.66 60.22 56.14 65.58

This Fig.5 shows the estimated output of FKP image for 0.1 67.76 74.73 63.45 72.89
the pre-processing step that includes the ROI and enhanced 1 73.21 80.40 77.11 81.13
image using histogram equalization

978-1-5386-4552-9/18/$31.00 ©2018 IEEE


2018 International Conference on Soft-computing and Network Security (ICSNS)
[4] Imed Bouchrika,Ammar Ladjaiclia,"Automated Clustering And
Estimation Of Age Groups From The Face Images Using Local Binary
Pattern operator",4th International confernce on Electrical
Engineering,2015,pp:1-4
[5] A.Deepa,T.Sasi Praba,"Age Estimation In Facial Image Using
Histogram Equalization",8 th International Conference On Advanced
Computing(ICOAC)2016,pp:186-190
[6] Konstantious G.Desparis,"Gabor filters",York University,2007,pp:1-5
[7] Tapan kunnar Harza,"Optical character recognition using KNN on
custom dataset",8 th Annual Industrial And Elctromechanical
Engineering Conference(IEMECON),pp:110-114,2017
[8] Jian Hou,"Feature combination and the KNN framework on object
classification ",IEEE Transactions on neural networks and learning
systems,Vol.27,No.6,June 2016,pp:1368-1378

Fig. 8. GAR&FAR In age estimation

This fig.8 shows the performance of matching rate in case


of trained samples. Various FKP index and middle finger
dataset are verified to better GAR ratio.
V. CONCLUSION
In this work, a innovative method of age estimation from
FKP based on Gabor filter &KNN is trained and samples are
tested .This Gabor filter helps to extract the FKP features in
order to get the real and magnitude parts. Similarly KNN
algorithm also classifies the dataset based on trained values.
Exact estimation of age for various index & middle is nearly
about (20-60) years. This proposed work gives the estimated
success rate of age from (80.40-81.13) for Left and right
middle FKP dataset. Likewise Index fingers also have success
rate nearly to (73.21-77.11).More accuracy rate of age
estimation can be achieved for several number of samples in
each category. For efficient results of age estimation collection
of real time samples in each category is initiated. However it
is aimed to check for several other classifier in case best
success rate
ACKNOWLEDGMENT
We would like to be grateful for International Research
Center of Kalasalingam University for granting financial
assistance under the scheme of University Research
Fellowship (URF) and we also acknowledge the department of
Electronics and Communication Engineering of Kalasalingam
University, (Kalasalingam Academy of Research and
Education), Tamil Nadu ,India for allowing to use the
computational facilities available in Signal Processing and
VLSI Design laboratory which was setup with the assistance
of the Department of Science and Technology (DST)

REFERENCES
[1] Hu han,Anil k.Jain,"Age ,Gender and Race estimation from
unconstrained Face Images",Msu Technical Report (2014), Pp;1-9
[2] Gnansivam.P,Dr.s Muttan,"Estimation of Age Through Fingerprints
Using Wavelet Transform And Singular Value Decomposition",
International journal of biometrics and bioinformatics, Vol(6);
Issue.1,2012.
[3] Ali Abdul Mun,im ibrahim,"Iris recognition using Gabor filters",Al
Tanqani,Vol.21,No.7,2008

978-1-5386-4552-9/18/$31.00 ©2018 IEEE

You might also like