Introduction To Palmprint Recognition

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1946
INTRODUCTION TO PALMPRINT RECOGNITION
Kanchana .A 1, Stanly Jayaprakash .J2
1 Assistant Professor, Head of the deprament of CSE,
Mahendra Engineering College for Women, Namakkal Dt
2 Assistant Professor, Head of the deprament of CSE,
Mahendra Institute of Technology, Namakkal Dt
----------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Biometrics refers to human characteristics
and behavior. Biometric authentication is used for
identification and access control that are used to identify
individuals in groups under observation. Biometric
identifier is frequently classified as physiological versus
behavioral characteristics, for examples, fingerprint, palm
veins, face recognition and palmprint. Existing works have
used multiple correlation filters per class for performing
palmprint classification algorithm. It produces sharp peak
for known class and noisy output for unknown sample
class using images from the PolyU database.
1. INTRODUCTION TO BIOMETRICS
Biometrics refers to metrics related to human a
characteristic which is used to identify individuals in
groups. Biometric systems resolve or verify a person’s
uniqueness without human intervention. It is based on
anatomical and behavioral traits such as fingerprint,
palmprint, vein pattern, finger knuckles, face, Iris, voice,
and gait. Biometric characteristics, also called as
templates, indicate a robust and constant
communication between a person and his identity and
these characteristics cannot be easily misplaced or
forgotten or distributed or forged. A biometric
characteristic called the feature set is exceptional for
each and every individual. The feature set obtained
during employment is stored in the system database as a
template.
2.INTRODUCTION TO PALMPRINT RECOGNITION
In recent years, a range of automated biometric
based recognition methods has been developed such as
fingerprint, palmprint, iris pattern, voice, etc. In these
biometric features, palmprint has a superior similarity
over the fingerprint. Palmprint is an image obtained
from the palm region of the hand. Palmprint based
systems have become fresh research areas in recent
years. The main benefit of palmprint is the accessibility
of large space for extracting biometric features. In
general, palmprint images should be standardized and
oriented before feature extraction. Palmprint devices are
less expensive than the other devices like iris.
Palmprints hold extra distinctive features such as
wrinkles and principle lines that can extract from low-
resolution images. By joining all features of palm and
fingerprint such as ridge and valley features, principle
lines and wrinkles, it is possible to provide a highly
accurate biometric system. Palm and finger biometrics
denoted by the information is presented in a friction
ridge impression. This information uses ridge flow, ridge
traits, and ridge structure of the raised segment of the
epidermis.
Palmprints and Fingerprints have both individuality and
stability. They have been used over a century as trusted
forms of identification. The general framework of
palmprints recognition system is depicted in figure 1.
Figure – 1: A framework for Palmprint recognition
Note that there is an extra point localizer of
wrinkles and principle lines in palmprint recognition
system. To Identify the accurate position of the lines for
palmprint recognition, the developer usually exploits the
locations of both lines and marks for help. In this system,
functional modules such as palm scanner, line localizer,
and identifier track the feature extractor as well as
pattern recognizer schematics.
2.1 Human Identification using Palm Vein Images
Yingbo Zhou & Ajay Kumar (2011) presented a human
identification system using palm vein images that consist
of two new approaches to enhance the operation of palm

vein-based identification systems. The palm vein-based
identification approach contains the possible
deformations, revolving, and translational variations by
training the orientation maintaining features and
utilizing a novel region-based matching scheme. There
are two types of approaches in evaluating of palm vein
that are obtained with the contactless and touch-based
imaging system. Palm vein images in contactless imaging
provide many translational and rotational changes.
Hence, more severe preprocessing steps are needed and
remove a constant and aligned ROI. This ROI reduces the
rotational, translational and scale variations. The
preprocessing steps fundamentally improve a fixed-size
ROI from the obtained images. This is pursued by the
nonlinear enhancement with the intention that the vein
patterns from ROI images are examined more obviously.
The identification process is more useful and efficient
and it is necessary to build a coordinate system. It is
sensible to link the coordinate system with the palm
itself. Two links are utilized as the orientation
points/lines to construct the coordinate system, i.e. the
link between the index finger and middle finger together
with the links between the ring finger and little finger.
Figure - 2: Block diagram for personal identification
using palm vein images
Figure 2 depicts personal identification using palm vein
images. The palm images are binarized and it extracts
the region of palm from the background region. Next, the
calculation of the distance from the center position of the
binarized palm to the boundary of the palm is done. The
system places the two links by observing the
corresponding local minima from the forecast distance.
The possible scale variations in the contactless
environment can be quite huge. It selects the location
and size of the ROI consistent with certain image-specific
calculations from the palm. But the method is more
computationally well-organized as no extra
sampling/computations are needed. After segmentation,
the ROI images are scaled to produce a constant size
region and the process is illustrated in Figure 3.
Figure - 3: Major steps in segmenting ROI images from
contactless palm vein images
The palm vein images employed in identification
using palm vein obtained under near-infrared
illumination (NIR) and the images normally emerge
darker with low contrast. So, image development more
evidently demonstrates the vein and texture patterns.
First, approximate the condition intensity profiles by
segmenting the image into a little overlapping. In 32 X 32
blocks, three pixels are overlapped between two blocks
to address the blocky effect. The average gray-level
pixels in each block are forecasted. Consequently, the
estimated background intensity profile is reduced to the
same size as the unique image using bicubic interruption
and the resulting image is subtracted from the original
ROI image. Finally, histogram equalization is adapted to
obtain the normalized and improved palm vein image.
The improvement has been moderately successful in
enhancing the information and contrast of the ROI
images.
The standardized and improved palm vein images
represent curved vascular network/patterns which can
be approximated by small line segments that are quite
curved. Consequently, in this work, a recognition
technique is developed to use two new approaches for
extracting such palm vein features. Additionally, an
adjacent matching scheme is provided that can
successfully report more regular, rotational and
translational changes and also to some extent image
deformation is obtained an image.

2.2 Palmprint Authentication using Fast Complex
Gabor Wavelet
Palmprint biometric is acquired for personal
authentication as it is distinctive and moderately low-
resolution images of less than 100 dpi are adequate to
remove the unique features. Palmprint features include
line, geometry, point, texture, and statistical features.
Jyoti Malik et al. (2011) observed that line features are
extracted by using complex Gabor wavelet transform
method. The line features extracted by complex Gabor
wavelet and the values of theta are loaded in the feature
vector. The feature vector is matched by hamming
distance similarity measurement using sliding window
method. In general, the palmprint authentication system
is divided into two subsystems:
(i) The pre- authentication system and
(ii) Authentication system
In the pre-authentication system, a database of Gabor
palmprint features is maintained. Reference threshold
values are also identified and stored in the database.
Figure – 4: Palmprint Pre-Authentication system
The above figure4 shows palm pre-authentication
systems. In this system, the accuracy of a person being
real or an imposter is identified with the aid of reference
threshold values stored in a pre-authentication system
database.
Figure – 5: Palmprint Authentication Systems
Figure 5 illustrated palmprint authentication system that
plays a significant function in making the real-time
authentication system. The reduction in matching time is
done in a proposed fast palmprint authentication system
using sliding window methods.
The desired line features are extracted from the
palmprint using complex Gabor wavelet method that
extracts line features from the input palmprint image.
The Gabor wavelet is basically a Gaussian modulated by
a complex sinusoid. The value of hamming distance is
forecast scaling value by using sliding window method.
In sliding window method, the ROI is minimized by the
window size (WS) and the window of ((60–WS)×(60–
WS)) slides over the rows and columns. The minimum
value of the Hamming distance values is considered if
the Hamming distance value is lower than the reference
threshold value.
The sliding window method, the vital approach of
feature matching, is an efficient method but it is very
slow due to the window size. The window size is selected
in such a way to negotiate association problem in the
palmprint images. Here, two sliding window approaches
are developed. Sliding window method 1 (SWM1) and
Sliding window method 2 (SWM2) are about using
palmprint segment in such a way to minimize the
matching operation time. Finally, the SWM1 and SWM2
reduce the complexity of the algorithm of SWM.
3. PALMPRINT TEXTURE VERIFICATION
In palmprint approach, images are convolved with one
Gabor filter. For each place in the region of interest
(there are 32 × 32 block locations), Gabor reaction is
modified to a binary format. It can be taken into account
as a feature reduction method, as Gabor reaction will be
1 or 0, followed by Hamming distance used as a
classifier. This approach, segmented as texture based on
Gabor filters, is frequently used as texture discriminator.
Since these filters can replicate lines adequately, the
following methods are used to detect line orientation.

The main parts of palmprint verification system are
depicted in Figure. 6.
Figure – 6: Main modules of a Palmprint Verification
The main modules of a palmprint verification system as
depicted in Figure 6 are:
a) Palmprint senses the palmprint of a human
obtained by a palmprint scanner.
b) Preprocessing in which the input palmprint is
improved and adapted to shorten the process of
feature extraction;
c) Feature extraction in which the palmprint is
further processed to produce discriminative
features also called feature vectors
d) Matching in which the feature vector of the
input palmprint compared against with one or
more existing templates.
The templates of supported clients of the
biometric system also called users are regularly
registered in a large database. Users can emphasize an
identity and their palmprints can be checked against
stored palmprints. Texture information is detected using
Haralick’s features on principle line boundaries. Texture
related palmprint recognition approach is used in a large
range of problems, such as cancer characterization in
medical imaging or segmentation of urban areas in
satellite and aerial images. Therefore, as per recent
studies, Haralick’s features are performed in higher
numbers for palmprint recognition.
In a training stage, paradigm textures are recognized in a
set of images representative of different classes. These
paradigm textures form a thesaurus of textons. A
palmprint is indicated by a histogram of the frequencies
of each text on. This idea must be used for other features
rather than texture, and the dictionaries are called
collections of visual words. Further, detailed texture
based palmprint recognition is explained in the sub-
sections below.
3.1 Reliable Contactless Palmprint Authentication
Aythami Morales et al. (2010) discussed the method for
automatic hand recognition. The work proposed a new
method for contactless hand authentication in complex
images. Contactless hand authentication system uses
skin color and hand shape information for hand
detection process. Next, the palm is extracted and
characterized by circular Gabor filter. Finally, the palm
features are matched by a new normalized approximated
string matching.
Typically the skin color is modeled by Bayes
classifiers or Gaussian mixture models. As an alternative
to Bayes classifiers and Gaussian mixture methods, the
skin color based model is designed for machine learning.
For an optimal compromise between the execution time
and the precision of detection, the system uses a neural
network (NN). And the NN entries are compiled by three
neurons. One is for the color component of pixels in RGB
domain. The NN output is the chance that a pixel is a skin
pixel. The learning phase allows modeling the skin tone
in RGB domain. Similarly, a Principal Components
Analysis on a skin pixels database identifies an exact
color space named skin space. After the learning, the NN
can notice the pixels looking like skin. For each pixel of
an image, the NN forecasts the chance that each pixel is a
skin pixel. This process constructs the probabilities map.
The segmentation by skin color cannot perform the
process of hand detection in a strong manner. Therefore,
a precise active shape model is defined to cancel this
problem and to address the two major complexities of
active shape model. These problems are the contour
initialization in the detection phase. After hand
detection, it is efficient to describe the palm separately of
the distance between the hand and the capture device.
The extraction is based on hand dimensions and the
palm extraction method.
The Gabor filter is implemented in texture
analysis by using Gaussian modulated by a wave. In
oriented Gabor filters, wave is an oriented complex
sinusoidal signal. This forceful circular Gabor filter is
convoluted with the palm image to artifact the palmprint
features. In contrast, the features are not binarized, so all
the information is maintained properly. With all
information, the changes between two data’s are
conserved.
Approximated String Matching Algorithm (AMSA) is
based on Levenshtein distance that is a simplification of
Hamming distance. At the start, the AMSA allows to
evaluate two vectors in forecasting a scoring matrix with
a scoring function. This function evaluates the similarity
between two components. The matching score is the

high in the scoring matrix. This scoring matrix contains
high matching score. Finally, the AMSA is strong to
translation, deletion, substitution or addition of elements
in the vectors.
3.2 Palmprint Biometric Template Security
(i) Koen Simoens et al. (2012) presented a system
model in which each center entity or grouping of
entities is a feasible attacker. Also, the image is
moldered by the differential morphological
pyramid, causing a separate scale-space
representation.
Major reasons for the necessity of Renewable
Biometrics: Firstly, biometric properties are subject to
biological and mechanical changes such as aging, and
injuries; hence the accuracy of the biometric
authentication may decrease over time. More
specifically, for behavioral biometrics such as speech or
handwriting, it is relatively understandable that aging
collides with the way that people talk.
From the point of biometric systems, this observation
results in the inclination of a possible increase of false
non-matches, i.e. legitimate users of biometric systems
are more often rejected that the effect has been
addressed. Secondly, cooperation or stolen biometric
data are complex for biometric systems. Once any unique
biometric raw data has been compromised, it may be
efficiently used for replay attacks. For instance, it has
been shown that sticky palmprints can be generated
from digital palmprint images and used for illegal gain
authentication by palmprint systems. For both the
reasons, it may be attractive to restore biometric
reference data: one objective is to preserve the
identification performance for individual subjects over
time of the process of biometric systems, by often
updating reference data.
4. PALMPRINT PATTERN MATCHING
The palmprint pattern mainly comprises of ridges and
minutiae, similar to a fingerprint pattern. Though, the
creases and ridges in palmprint regularly overlap each
other.
Therefore, reports have the extraction of local palmprint
features. But, most of the existing work is limited only to
the extraction of ridges.
Nevertheless, a few works attempted to estimate
palmprint crease points by generating a local gray level
directional map. These crease points are linked together
to separate the crease in the form of line segments,
which are used in the matching task. No features are
offered to recommend the strength of these partly
extracted creases for the matching of palmprints. Some
related works on palmprint verification are listed in the
sub-sections.
4.1 Palmprint Recognition using Image Patterns
using Correlation Filters
Vishnu Naresh Boddeti and BVK Vijaya Kumar., (2013)
presented a template-based framework to bind class-
specific information to a set of image patterns. It
recovers that information by matching the template, and
a query pattern of the same class must describe. This is
performed by mapping the class-specific information to
a group of spatial.
The information is retrieved during matching with a
genuine query by forecasting the spatial translations
applied to the images that were used to plan the
template.
The main concentration made on the problem of binding
information to biometric marks as an application of the
framework. The main concept behind the approach is to
exploit Correlation Filters (CFs) for the dual reason of
pattern matching and binding specific class information
to the template.
During authentication, the bound information is released
automatically if the query pattern is authentic. The
enrollment and authentication stages of the CF related
framework are described in Figure 7.
Figure – 7: Enrollment and Authentication Stages of CF
related Framework
In this stage, the following points are required.
 Training images representative of the
authentic class. These could be biometric
signatures under various conditions

(eg. face images with different lighting,
expressions etc.) expected during testing.
 Information or key to bound to the template
 A secondary input (e.g., password or pin)
may be required for additional protection
The sequence should be mapped into places in the
correlation plane, by segmenting the key into smaller
segments of suitable size. Perfectly, the multi-peak CF
should generate correlation peaks at those places in
reaction to a centered valid image. For an un-centered
query pattern, the peaks transfer internationally
depending on the unidentified qualified shift between
the query and the training images. Consequently during
training, the system forecasts the centroid of the
specified peak places that are stored in the database.
During authentication, once the peaks should identify,
they shift the centroid of the detected peaks to the region
stored in the database. Additionally, as the key is
mapped only to regions in the correlation plane, it can be
recovered only up to a variation if the order of the key
segments is not encoded. While other patterns are
possible, the work resolves this issue by encoding the
ordering along with the key in the filter itself by
augmenting the key with the order of that segment.
Figure 8 shows an example of mapping the binary key to
peak location.
Figure – 8: Example showing mapping from binary key
to peak location
The multipeak CF is designed with the training images,
and the constraints are obtained from the key as inputs.
This template is stored in the database along with the
hash value of the key (this is optional) computed using a
one-way hash function. Further, the work also does not
allow the regions of the constraints to be within 3 pixels
from the boundary of the image.
In an authentication stage, the query pattern is offered
along with secondary input and a claimed identity in a
verification scenario. Figure 9 shows a block diagram of
the authentication process.
Figure -9: Block diagram of the testing stage in
authentication
The query is then cross-correlated using the CF matching
to the claimed identity. If the query is genuine, the
resulting correlation should have peaks at the right
places except for a possible global shift. The detected
peaks in the center portion of the palmprint are to be
stored in the database. From the new peak locations, the
information bound to the template is reconstructed. The
correctness of the retrieved key can be confirmed by
comparing the stored hash value with the hash value of
the recovered information. For a fake query image, key
recovery would be unsuccessful due to the lack of sharp
peaks in the resulting correlation plane when either the
secondary input is incorrect, or the query does not
belong to the claimed class or when both are incorrect.
Thus, the framework to bind information to image
patterns and to retrieve this information during
authentication by embedding the information in the
template is designed to discriminate that pattern class
from the other pattern classes.
4.2 Multi feature Based Palmprint Recognition
Jing Claim & Jie Zhou (2011) observed that multi feature
based high resolution palmprint recognition improves
the matching performance with dissimilar resolution
that exists in the contactless database. Quality-based and
adaptive orientation field estimation algorithm with
Neyman-Pearson rule could achieve the good
performance, yet non-linear deformation and matching
efficiency are not up to the grade.
The main contributions of the multi feature based high
resolution palmprint recognition include the following:
Prob
e
Hash key
Retrieve
Key
Correlation
Autho
r
Release
Key
Hash key
Key
Image
Transform
ation
Secondary
Input

1) Utilize multiple features, namely minu tiae,
solidity, orientation and principle lines for palmprint
recognition to considerable progress the matching
performance of the conventional algorithm.
2) Propose quality-based and adaptive
orientation field evaluation algorithm that executes
better than the existing algorithm in case of regions
with a large number of creases.
3) Exploit fresh fusion scheme for a
recognition application that performs better than
predictable fusion methods, e.g. weighted sum rule,
SVMs or Neyman-Pearson rule.
The work evaluates the discriminative energy of
different feature fusions and finds concreteness is very
useful for palmprint recognition. An analysis of the
database containing 14,576 full palmprints proved that
the proposed algorithm achieved optimal results. In the
verification process, False Rejection Rate (FRR) is 16%
and 17% lower than the existing algorithm. It is at a False
Acceptance Rate (FAR) of 10%. In identification process,
the rank-1 live-scan partial palmprint recognition rate is
improved from 82 % to 91.7 %.
Line feature matching is reported to be dominant and
provides high accuracy in palmprint verification. On the
other hand, it is very complex to describe these palm
lines, i.e. their magnitude and direction, in noisy images
precisely.
With the motivations of these existing works in
palmprint texture recognition, palmprint pattern
matching, and palmprint template recognition, the
proposed work is designed in an efficient way. And also
by considering the limitation faced in the existing work,
the proposed work is designed to overcome those
difficulties.
5. CONCLUSION
Palmprint is the accessibility of large space for
extracting biometric features. In general, palmprint
images should be standardized and oriented before
feature extraction. Palmprints hold extra distinctive
features such as wrinkles and principle lines that can
extract from low-resolution images. By joining all
features of palm and fingerprint such as ridge and valley
features, principle lines and wrinkles, it is possible to
provide a highly accurate biometric system. Palm and
finger biometrics denoted by the information is
presented in a friction ridge impression.
REFERENCES
[1] Adams Kong., David Zhang & Mohamed Kamel
2009, ‘A Survey on Palmprint Recognition’, Pattern
Recognition Letters, ELSEVIER, and vol. 42, no. 7,
pp. 1408-1418.
[2] Adebayo Daramola, S, Olujimi Ajayi & Tiwalade Od
2014, ‘Robust Palm-Print Feature for Human
Identification’, International Journal of Engineering
and Advanced Technology (IJEAT), vol. 3, no. 4,
pp. 152-155.
[3] Ajay Kumar & David Zhang 2006, ‘Personal
Recognition Using Hand Shape and Texture,” IEEE
Transactions on Image Processing, vol. 15, no: 8, pp.
2454-2461.
[4] Ajay Kumar, David Wong, CM, Helen Shen, C & Anil
Jain, K 2003, ‘Personal Verification using Palmprint
and Hand Geometry Biometric’, Audio and video
based biometric person authentication, Springer,
vol. 2688, pp.no:668-678.
[5] Anu Nair, SH & Aruna, P 2014, ‘Image Fusion
Techniques for Iris and Palmprint Biometric
System’, Journal of Theoretical and Applied
Information Technology, vol. 61, no. 3, pp. 476-485.
[6] Arie Nakhmani & Allen Tannenbaum 2012, ‘Self-
Crossing Detection and Location for Parametric
Active Contours’, IEEE Transactions on Image
Processing, vol. 21, no. 7, pp. 3150-3156.
[7] Ashutosh Kumar & Ranjan Parekh 2012, ‘Palmprint
Recognition in Eigen-space’, International Journal
on Computer Science and Engineering (IJCSE).
[8] Aythami Morales, Miguel FEERer, A & Ajay Kumar
2010, ‘Improved Palmprint Authentication using
Contactless Imaging’, IEEE Transactions on
Information Forensics and Security.
[9] Benjamin Snyder & Regina Barzilay 2008,
‘Unsupervised Multilingual Learning for
Morphological Segmentation’, Proceedings of ACL-
08: HLT, pp. 737–745.
[10] Cappelli, R, FEERara, M & Maio, D 2012, ‘A Fast and
Accurate Palmprint Recognition System Based on
Minutiae’, IEEE Transactions on Systems, Man, and
Cybernetics, vol. 42, no. 3, pp. 956-962.
[11] Cuong Vu, T, Thien Phan, D & Damon Chandler, M
2012, ‘S3: A Spectral and Spatial Measure of Local
Perceived Sharpness in Natural Images’, IEEE
Transactions on Image Processing, vol. 21, no. 3, pp
1-13.
[12] Dewi Yanti Liliana & Eries Tri Utaminingsih 2012,
‘The combination of palmprint and hand geometry
for biometrics palm recognition’, International
Journal of Video & Image Processing and Network
Security IJVIPNS-IJENS vol. 12, no.1, pp. 1-5.

Introduction To Palmprint Recognition

More Related Content

What's hot (20)

Similar to Introduction To Palmprint Recognition (20)

More from IRJET Journal (20)

Recently uploaded (20)

Introduction To Palmprint Recognition