Estimation of Age Through Fingerprints Using Wavelet Transform and Singular Value Decomposition

P. Gnanasivam & Dr. S. Muttan
International Journal of Biometrics and Bioinformatics (IJBB), Volume (6) : Issue (2) : 2012 58
Estimation of Age Through Fingerprints Using Wavelet
Transform and Singular Value Decomposition
P. Gnanasivam pgnanasivam@yahoo.com
Department of ECE
Agni College of Technology
Chennai, 603103, India
Dr. S. Muttan muthan_s@annauniv.edu
address.com
Centre for Medical Electronics, Department of ECE,
College of Engineering
Anna University
Chennai, 600025, India
Abstract
The forensic investigators always search for fingerprint evidence which is seen as one of the
best types of physical evidence linking a suspect to the crime. In this paper discrete wavelet
transform (DWT) and the singular value decomposition (SVD) has been used to estimate a
person’s age using his/her fingerprint. The most robust K nearest neighbor (KNN) used as a
classifier. The evaluation of the system is carried on using internal database of 3570
fingerprints in which 1980 were male fingerprints and 1590 were female fingerprints. Tested
fingerprint is grouped into any one of the following five groups: up to 12, 13-19, 20-25, 26-35
and 36 and above. By the proposed method, fingerprints were classified accurately by
96.67%, 71.75%, 86.26%, 76.39% and 53.14% in five groups respectively for male and by
66.67%, 63.64%, 76.77%, 72.41% and 16.79% for female. Finger-wise and Hand-wise results
of age estimation also achieved.
Keywords: Age Estimation, Discrete Wavelet Transform, Singular Value Decomposition, K-
Nearest Neighbor.
1. INTRODUCTION
Age information is important to provide investigative leads for finding unknown persons.
Existing methods for age estimation have limited use for crime scene investigation because
they depend on the availability of teeth, bones, or other identifiable body parts having physical
features that allow age estimation by conventional methods. In this paper, age of a person is
estimated from the fingerprints using DWT and SVD. The science of fingerprint has been
used generally for the identification or verification of person and for official documentation.
Fingerprint analysis plays a role in convicting the person responsible for an audacious crime.
Fingerprint has been used as a biometric for the gender and age identification because of its
unique nature and do not change throughout the life of an individual [1].
In fingerprint, the primary dermal ridges (ridge counts) are formed during the gestational
weeks 12-19 and the resulting fingerprint ridge configuration (fingerprint) is fixed permanently
[2-3]. Ridges and their patterns exhibit number of properties that reflect the biology of
individuals. Fingerprints are static and its size and shape changes may vary with age but
basic pattern of the fingerprint remains unchanged. Also, the variability of epidermal ridge
breadth in humans is substantial [4]. Dermatoglyphic features statistically differ between the
sexes, ethnic groups and age categories [5]. Gender and age determination of unknown can
guide investigators to the correct identity among the large number of possible matches.
According to ‘Police-reported crime statistics in Canada, 2010 [6], Crimes tend to be
disproportionately committed by youth and young adults and the rate of those accused of
a Criminal Code offence peaked at 18 years of age and generally decreased with increasing
age. As per the publication of ‘crime in India statistics-2010, published by National Crime

Records Bureau [7], the crime rate is higher for the age range of 18 to 44 and decreases after
44. Thus the age estimated is categorized in to any one of the five groups: up to 12, 13-19,
20-25, 26-35 and 36 and above. The features of the fingerprints grouped are trained and
classified using the KNN classifier. Figure 1 illustrates the process of DWT and SVD based
age estimation method.
FIGURE 1: DWT and SVD based age estimation method
Only few works were concentrated in the age estimation using the fingerprint. In the view of
protecting children from accessing online sites that are harmful to minors, John D Woodward,
testified before a hearing of ‘commission on online child protection’ [8]. According to him,
there are no age verification biometrics, no age determination biometrics and no age
estimation biometrics. Based on the fingerprint ridge width the age and sex were determined
[4]. Many Earlier works were related the fingerprint image quality and the age of a person [9-
11]. Other works in the way of age estimation used the speech recognition [12], face [13-15]
etc. In this work, authors proposed a method of identifying range of the age using the discrete
wavelet transform and the singular value decomposition
Wavelet transform is a popular tool in image processing and computer vision because of its
complete theoretical framework, the great flexibility for choosing bases and the low
computational complexity [16]. As wavelet features has been popularized by the research
community for wide range of applications including fingerprint recognition, face recognition
and gender identification using face, authors have confirmed the efficiency of the DWT
approach for the gender identification using fingerprint.
The SVD approach is selected for the gender discrimination because of its good information
packing characteristics and potential strengths in demonstrating results. The SVD method is
considered as an information-oriented technique since it uses principal components analysis
procedures (PCA), a form of factor analysis, to concentrate information before examining the
primary analytic issues of interest [17]. K-nearest neighbours (KNN), gives very strong
consistent results. It uses the database which was generated in the learning stage of the
proposed system and it classifies genders of the fingerprints.
The outline of this paper is as follows: the fingerprint feature extraction using DWT and SVD
is described in Section 2; we then proposed the age classification using fingerprint features in
Section 3; the experimental results are presented in Section 4; Section 5 comes to the
conclusion and future work.
2. FINGERPRINT FEATURE EXTRACTION
Feature extraction is a fundamental pre-processing step for pattern recognition and machine
learning problems. In the proposed method, the energy of all DWT sub-bands and non-zero
singular values obtained from the SVD of fingerprint image are used as features for the
estimation of age. In this section, DWT and SVD based fingerprint feature extractions are
described.
2.1 DWT Based Fingerprint Feature Extraction
Wavelets have been used frequently in image processing and used for feature extraction, de-
noising, compression, face recognition, and image super-resolution. Two dimensional DWT
Frequency
Feature
Extraction using
DWT
DWT Feature
vector
SVD Feature
vector
Spatial Feature
Extraction using
SVD
Feature
Fusion
KNN
Classifier
for Age
Estimation
D
F E
I C
N I
A S
L I
O
N

decomposes an image into sub-bands that are localized in frequency and orientation. The
decomposition of images into different frequency ranges permits the isolation of the frequency
components introduced by “intrinsic deformations” or “extrinsic factors” into certain sub-
bands. This process results in isolating small changes in an image mainly in high frequency
sub-band images. Hence, discrete wavelet transform (DWT) is a suitable tool to be used for
designing a classification system.
The 2-D wavelet decomposition of an image is results in four decomposed sub-band images
referred to as low–low (LL), low–high (LH), high–low (HL), and high–high (HH). Each of these
sub-bands represents different image properties. Typically, most of the energy in images is in
the low frequencies and hence decomposition is generally repeated on the LL sub band only
(dyadic decomposition). For k level DWT, there are (3*k) + 1 sub-bands available. The
energy of all the sub-band coefficients is used as feature vectors individually which is called
as sub-band energy vector (E). The energy of each sub-band is calculated by using the
equation (1).
∑∑= =
=
R
i
C
j
kK jix
RC
E
1 1
),(
1
(1)
Where xk (I, j) is the pixel value of k
th
sub-band and R, C is width and height of the sub-band
respectively.
Figure 2 shows the block diagram of the frequency feature extraction by using DWT. The
input fingerprint image is first cropped and then decomposed by using the DWT. For level 1,
number of subbands are 4 and 3 subbands are added for each next levels Thus the increase
in levels of DWT increases the features.
FIGURE 2: DWT based fingerprint feature extraction
2.2 SVD Based Fingerprint Feature Extraction
The Singular Value Decomposition (SVD) is an algebraic technique for factoring any
rectangular matrix into the product of three other matrices. Mathematically and historically, it
is closely related to Principal Components Analysis (PCA). In addition it provides insight into
the geometric interpretation of PCA. As noted previously, the SVD has long been considered
fundamental to the understanding of PCA.
The SVD is the factorization of any (k x p) matrix into three matrices, each of which has
important properties. That is, any rectangular matrix A of k rows by p columns can be factored
into U, S and V by using the equation (2).
T
USVA = (2)
Where
T
AAU = (3)
AAV T
= (4)
And S is a (k x p) diagonal matrix with r non-zero singular values on the diagonal, where r is
the rank of A. Each singular value is the square root of one of the Eigen values of both AA
T
and A
T
A. The singular values are ordered so that the largest singular values are at the top left
and the smallest singular values are at the bottom right, i.e., s1,1≥s2,2≥s3,3 etc. Among the
three rectangular matrices, S is a diagonal matrix which contains the square root Eigen
values from U or V in descending order. These values are stored in a vector called Eigen
vector (V). As the internal database contains images of size 260x300 pixels, the feature
Training
Sample
Cropping Discrete Wavelet
Transform
Extraction of
Sub-band Energy
Vector (Ek)

vector of SVD is of the size 1x260. The spatial feature extraction by using SVD is shown in
Fig 3.
FIGURE 3: SVD based fingerprint feature extraction
3. FINGERPRINT AGE ESTIMATION
The proposed system for age estimation is built based on the concatenation of fingerprint
features obtained by using DWT and SVD. This section describes two different stages named
as learning stage and classification stage and the KNN classifier used for the age
classification.
4.1. Learning Stage
The feature vector V of size 1x260 obtained by SVD and the sub band energy vector E of size
1x19 obtained by DWT are fused to form the feature vector and used in the learning stage.
The fusion of feature vector V and E is done by concatenation of features that are widely used
for feature level fusion. The resulting feature vector is of the size 1x279 (1x260 +1x19).The
learning stage is shown in Fig 4.
FIGURE 4: Learning stage of the proposed gender classification system
The learning algorithm is as follows:
Learning Algorithm:
[Input] all samples of fingerprint with known class (Gender)
[Output] the feature vector of all samples as database
1) Decompose the fingerprint with 6 level decomposition of DWT.
2) Calculate the sub-band energy vector (E) using (1).
3) Calculate the Eigen vector (V) using (2).
4) Fuse the vectors E and V to form the feature vector for the particular
fingerprint.
5) Insert this feature vector and the known class into the database.
6) Repeat the above steps for all the samples.
4.2. KNN Classifier
In pattern recognition, the k-nearest neighbour algorithm (K-NN) is the generally used method
for classifying objects based on closest training examples in the feature space. K-NN is a type
of instance-based learning where the function is only approximated locally and all
computation is deferred until classification. In K-NN, an object is classified by a majority vote
of its neighbours, with the object being assigned to the class most common amongst its k
nearest neighbours (k is a positive integer, typically small). If k = 1, then the object is simply
assigned to the class of its nearest neighbour. The neighbours are taken from a set of objects
for which the correct classification is known. This can be thought of as the training set for the
algorithm, though no explicit training step is required.
Singular Value
Decomposition
Extraction of
non-zero
Singular values
Training
Sample
Training
Sample
DATABASE
Frequency Features
Extraction
Spatial Features
Extraction
Feature
Fusion

4.3. Classification Stage
In the classification phase, the fused feature vector of the input fingerprint is compared with
the feature vectors in the database by using the K-Nearest Neighbour classifier. The distance
measure used in the classifier is ‘Euclidean Distance’. The detailed classification algorithm is
detailed below.
Classification Algorithm
[Input] unknown fingerprint and the feature database
[Output] the class of the fingerprint to which this unknown fingerprint is assigned
1) Decompose the given unknown fingerprint with 6 level decomposition of
DWT.
2) Calculate the sub-band energy vector (E) using (2).
3) Calculate the Eigen vector (V) using (1).
4) Fuse the vectors E and V to form the feature vector for the given
unknown fingerprint.
5) Apply KNN classifier and find the class of the unknown fingerprint by
using the database generated in the learning phase.
4. EXPERIMENTAL RESULTS
In this section, the performance of the proposed Age estimation algorithm is verified by using
the internal database. The success rate (in percentage) of age classification using the
combination of both DWT and SVD are summarized and discussed. DWT level 5, 6 and 7
were tried and from the results, DWT level 6 is identified as the optimum for the age
estimation.
4.1. Data Set
The fingerprint images of internal database were collected by using Fingkey Hamster II
scanner manufactured by Nitgen biometric solution [18], Korea. Every original image is of size
260x300 pixels with 256 grey levels and resolution of 500 dpi. The internal database includes
all ten fingers of the subject scanned. The images are collected from males and females of
different ages. From the internal database, irrespective of quality and age, all ten fingers of
3570 fingerprints in which 1980 were male fingerprints and 1590 were female fingerprints are
used for testing and training. These 3570 fingerprint images are separated into two sets. For
the learning stage 2/3 of total images are used. The remaining images are used in the
classification stage. Table 1 shows the age and gender wise samples of the internal
database. The crime rate recorded is high between the ages of 12 to 35. Below the age of 12
and above 36 the crime rate is few and thus large samples from 283 persons (156 male and
127 female) were used for testing the method. These samples are classified under the three
groups as 13-19, 20-25 and 26-35. Each person’s all fingers are scanned and frequency
features were identified through DWT and spatial features were identified through SVD.
Initially, KNN was applied for the feature set of DWT alone. Similarly, KNN was applied for the
feature set of SVD alone and the fused feature set of DWT and SVD.
Age Group Male Female Total
Up to 12 70 60 130
13-19 190 320 510
20-25 1050 680 1730
26-35 320 270 590
36 and above 350 260 610
Total Samples 1980 1590 3570
TABLE 1: Age and gender wise samples details
All the fingers of a person are scanned to test the proposed algorithm. The proposed method
is tested with all five fingers of the left and right hand. Thus to have the reference of the
fingers, the scanned fingers were numbered as follows. Little finger to thumb fingers of left

hand is numbered as 1 to 5. Thumb to little fingers of right hand is numbered as 6 to10 as
shown in figure 5.
FIGURE 5: Finger numbering
4.1 Age Estimation
In table 2 the success rates (in percentage) of age estimation for the male fingerprints are
tabulated. For the fingerprints of the male persons whose age lies below 12 years, the
success rate is achieved with 96.67%.In this category, other than the left hand thumb, the age
is estimated exactly for all fingers. The success rate in the age group of 20-25 is reasonably
good (86.26%) and thus useful for crime investigation, as this group crime rate is higher than
other groups. In addition, if there is availability of right thumb and right index finger, the
success is nearly 90%. Similarly the success rate for the remaining group is achieved as
71.75%, 76.39% and 53.14% for the age groups of 13-19, 26-35 and 36 and above
respectively. Maximum success rate is achieved in the age group of ‘up to 12’ for all fingers
except the left thumb. Low success rate is recognized for the age group of ’36 and above’.
Fingers
Age groups
up to 12 13-19 20-25 26-35
36 and
above
1 100.00 72.22 80.81 75.00 51.43
2 100.00 84.21 83.84 72.22 51.43
3 100.00 72.22 86.87 69.44 54.29
4 100.00 66.67 83.84 83.33 51.43
5 66.67 66.67 88.89 72.22 51.43
6 100.00 77.78 89.90 80.56 54.29
7 100.00 72.22 83.84 80.56 51.43
8 100.00 66.67 87.88 69.44 57.14
9 100.00 72.22 89.90 80.56 54.29
10 100.00 66.67 86.87 80.56 54.29
Average 96.67 71.75 86.26 76.39 53.14
TABLE 2: Success rate (in percentage) of age estimation for the male fingerprints
Similar to the male age estimation, the success rates (in percentage) of age estimation for the
female fingerprints are calculated. For the fingerprints of the female persons whose age lies
below 12 years, the success rate is achieved with 66.67%. The success rate in the age group
of 20-25 is reasonably good (76.77%) and thus useful for crime investigation, as this group
crime rate is higher than other groups. In addition, if there is availability of right and left small
fingers of female, the success is nearly 81%. Similarly the success rate for the remaining
group is achieves as 63.64%, 72.41% and 16.79% for the age groups of 13-19 26-35 and
above 36 respectively. Maximum success rate is recognized for the right little finger of the age

group ’20-25’ and the lowest success rate are notices in the right ringer finger of the age
group ’36 and above. Finger-wise age estimation success rate is tabulated in table 3.
Fingers
Age groups
up to
12
13-19 20-25 26-35
36 and
above
1 66.67 63.64 80.00 68.97 17.86
2 66.67 63.64 76.92 75.86 14.29
3 66.67 63.64 78.46 68.97 17.86
4 66.67 63.64 72.31 72.41 14.29
5 66.67 63.64 78.46 75.86 17.86
6 66.67 63.64 73.85 72.41 14.29
7 66.67 63.64 72.31 72.41 21.43
8 66.67 63.64 76.92 68.97 21.43
9 66.67 63.64 76.92 72.41 10.71
10 66.67 63.64 81.54 75.86 17.86
Average 66.67 63.64 76.77 72.41 16.79
TABLE 3: Success rate (in percentage) of age estimation for the female fingerprints
Age group-wise average success rate for male and female is shown in the line diagram of
figure 6. Maximum success rate of 96.67% is achieved for the age group of ‘up to 12’ years of
male fingers and 76.77% is achieved for ’20-25’ age group for female fingers. For the age
group of ‘36 and above’, the success rate is low and 53.14% and 16.79% for male and female
fingers respectively.
FIGURE 6: Age group-wise Average success rate
By average for any given fingers, 76.84% success rate is achieved for the male and 59.26%
is achieved for the female fingers in all the five groups. Also, age estimation for the right and
left hand are calculated. As discussed, left hand small finger to thumb fingers are named 1 to
5 and the right hand thumb finger to small finger are named 6 to 10. Average age estimation,
for the left hand fingers among all the fingers are calculated as 75.41% for male and 59.28%.
Similarly, 78.28% and 59.23% are the success rates for right hand fingers of male and female
respectively among all groups. Finger-wise success rate for the male and female fingers are
shown in figure 7(a) and 7(b) respectively.

60
65
70
75
80
85
1 2 3 4 5 6 7 8 9 10
75.9
78.3
76.6 77.1
69.2
80.5
77.6
76.2
79.4
77.7
SuccessRate(%)
Fingers
Male
(a)
50
55
60
65
1 2 3 4 5 6 7 8 9 10
59.4 59.5 59.1
57.9
60.5
58.2
59.3 59.5
58.1
61.1
SuccessRate(%)
Fingers
Female
(b)
FIGURE 7: Finger-wise success rate (a) Male (b) Female
5. CONCLUSIONS
In this work, we have proposed a new method of age estimation from fingerprint images
based on level 6 DWT and SVD. The level 6 DWT is selected as optimum level by analysing
the results obtained for other levels. DWT and SVD also applied to classify the fingerprints in
to the five age groups. According to the crime reports of Canada and Indian police record,
ages has been grouped as up to 12, 13-19, 20-25, 26-35 and above 35. Exact estimation of
age group is achieved for the age group below 12 years. For the age group of 36 and above
the success rate is not reasonably good.
Higher crime rate has been recorded in Canada and Indian police records in the ages, ranges
20 to 35 of both male and female. The proposed method gives age estimation rate of 81.33%
and 74.59 for male and female respectively. If the right thumb and right ring fingers of male is
given, the success rate is attained as 85.23% and 78.7% for female little fingers. The overall
success rate is 76.84% for male and 59.26% for female. While testing the right hand male
fingers alone, 78.28% of accurate estimation is attained. Similarly for the male, irrespective of
right or left hand the success rate is 59.3%. More accuracy rate of age estimation can be
achieved if more number of samples in each category is trained.

For better results of age estimation, authors are working in collecting huge samples in the
each category is initiated. In addition, the research work has been extended using the spatial
parameters of fingerprint. Moreover, it is aimed to use various other classifiers for the better
results. In the proposed work five age groups has been made for the age estimation. In the
future work, more groups will be made to find the age of the suspect or the person.
6. REFERENCES
[1] D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, “Handbook of Fingerprint
Recognition”, first ed., Springer, New York, 2003.
[2] J. John, Mulvihill, and David W. Smith, “The genesis of dermatoglyphics,”the journal of
Pediatrics, vol. 75, no. 4, pp. 579-589, 1969.
[3] W. Babler, “Embryologic development of epidermal ridges and their configurations,” In:
Plato CC, Garruto RM, Schaumann BA, editors. Dermatoglyphics: Science in Transition.
[4] Miroslav Kralik, Vladimir Novotny, “Epidermal Ridge Breadth: An Indicator of Age and
Sex in Paleodermatoglyphics”, Variability and Evolution, Vol. 11, pp. 5-30, 2003.
[5] Harold Cummins, Walter J. Walts, and James T McQuitty, “The breadths of epidermal
Ridges on the finger tips and palms - A study of variation.” American Journal of Anatomy,
vol. 68, no.1, pp. 127-150, 1941.
[6] Shannon Brennan and Mia Dauvergne, Police-reported crime statistics in Canada, 2010,
http://www.statcan.gc.ca/pub/85-002-x/2011001/article/11523-eng.htm
[7] Crime in India 2010-Statistics, National Crime Records Bureau, Ministry of Home Affairs,
Government of India, http://ncrb.nic.in.
[8] John D. Woodward, Jr., “Is Biometrics an age verification technology”, Testimony,
RAND’s publications, June 2000.
[9] Shimon K. Modi, Prof. Stephen J. Elliott, Jeff Whetsone and Prof. Hakil Kim,” Impact of
Age Groups on Fingerprint Recognition Performance” In IEEE Workshop on Automatic
Identification Advanced Technologies, 2007pp. 19-23.
[10] S. K. Modi, and S. J. Elliott, "Impact of image quality on performance: Comparison of
young and elderly fingerprints," in International Conference on Recent Advances in Soft
Computing (RASC), 2006 pp.10-12.
[11] S. J. Elliott, and N. C. Sickler, "An evaluation of fingerprint image quality across an
elderly population vis-a-vis an 18-25-year-old population," In International Conference
security Technology, 2005 pp. 68-73.
[12] Tobias Bocklet, Andreas Maier, Elmar Noth, “Age Determination of Children in Preschool
and Primary School Age with GMM-based Super vectors and Support Vector Machines/
Regression”, international conference on Text, Speech and Dialogue, 2008, pp. 1-9,.
[13] Mohammad EL Deeb and Motaz El-Saban, “Human age estimation using enhanced bio-
Inspired features (EBIF)”, In International Conference on Image Processing (ICIP), 2010,
pp. 1589-1592.
[14] Khoa Luu, Karl Ricanek, Tien D Bui, and Ching Y Suen, “Age estimation using Active
Appearance Models and Support Vector Machine regression”, in International
Conference on Biometrics Theory Applications and Systems, 2009, pp. 1-5.
[15] Andreas Lanitis, Chrisina Draganova, and Chris Christodoulou, “Comparing Different
Classifiers for Automatic Age Estimation”, IEEE Transactions on Systems, Man, and
Cybernetics-Part B: Cybernetics, Vol. 34, no.1, pp. 621-628, 2004.

[16] Bai-Ling Zhang, Haihong Zhang, and Shuzhi Sam Ge, “Face Recognition by Applying
Wavelet Subband Representation and Kernel Associative Memory”, IEEE Transactions
on neural networks, vol. 15, no. 1, pp. 166-177, 2004.
[17] G. Golub, and W. Kahane, “Calculating the singular values and pseudo-inverse of a
matrix”, Journal of Society for industrial and application mathematics series B: numerical
analysis, vol. 2, no. 2, pp. 205-224, 1965.
[18] Nitgen Company, Fingkey Hamster II fingerprint sensor http://www.nitgen.com/eng/

Estimation of Age Through Fingerprints Using Wavelet Transform and Singular Value Decomposition

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to Estimation of Age Through Fingerprints Using Wavelet Transform and Singular Value Decomposition (20)

Recently uploaded (20)

Estimation of Age Through Fingerprints Using Wavelet Transform and Singular Value Decomposition