Content uploaded by Saravanan Chandran
Author content
All content in this area was uploaded by Saravanan Chandran on Oct 10, 2014
Content may be subject to copyright.
1
Proc. of the Intl. Conf. on Advances In Computer and Electronics Technology- ACET 2014.
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-024-8 doi: 10.15224/ 978-1-63248-024-8-01
Performance Analysis of Edge Detectors for
Palmprint Matching
Dr. Saravanan Chandran, Ph.D.
Assistant Professor
National Institute of Technology, Durgapur, West Bengal, India, Pin – 713209
Dr.cs1973@gmail.com
Abstract— The palmprint has been used for future prediction of
human being. In the recent years the palmprint has been used for
biometric applications as human verification. The palmprint has
many lines of different sizes and directions. The lines are detected
as edges using popular edge detecting algorithms such as Sobel,
Prewitt, Roberts, Log, Zero-cross, and Canny. Thus, an analysis
work has been performed on these popular edge detection
algorithms to identify suitable edge detection algorithm which
improves the palmprint matching process. The experiment
results reveal that the canny edge detection algorithm identifies
complete set of edges of various sizes compared with other
popular edge detection algorithms. Moreover, the Sobel edge
detection algorithm identifies the medium and as well as longer
lines. Further, the Prewitt, Roberts, Log, and Zero-cross edge
detection algorithms ignore the small lines and identifies only the
main longer lines. Thus, the analysis work confirms the Canny
and Sobel edge detection algorithms are preferable for edge
detection algorithms.
Keywords— Palmprint, Edge Detection, Biometrics.
I. Introduction
Authenticating human beings at various places such as
Airport, Office, Internet payment, etc. is an essential task.
There are several techniques used such as punch card system,
magnetic swipe card system with pin number, fingerprint
verification system, palmprint verification system, face
recognition, etc. The failure rate of these systems are varies
from 1% to 5%. The system may have false acceptance or
false rejection or both. The failure of the verification system
may lead to security threat to the Airport or Office, etc.
The punch card system is outdated and in few places the
magnetic swipe card system is utilized. Some places the
fingerprint verification systems are implemented. Few places
the palmprint systems, face recognition systems are in use.
The palmprint is unique in features for a person. The
palmprint verification system is highly preferred due to the
simplicity of the system and speed. Since, the area of the
palmprint is higher than the fingerprint and lesser than the
face, is highly preferred for feature extraction. The palmprint
verification system has been implemented with the help of
touch less system using android base mobile phone [11].
II. Related Works
There are many researchers working palmprint verification
system. Few works are reported here.
L.J. Spreeuwers and F. Van Der Heijden proposed a new
method for evaluation of edge detectors based on the average
risk of a decision [1]. The average risk is a performance
measure well-known in Bayesian decision theory. They
described a method to estimate the probabilities on a number
of different types of errors. A weighted sum of these estimated
probabilities represented the average risk. The weight
coefficients defined the cost function. The method was
suitable, not only for the comparison of edge operators, but
also for the determining of the weaknesses and strengths of a
certain edge operator. They considered Sobel algorithm, the
Marr-Hildreth operator, and the Canny operator for
experiment. The experiment results that the Canny operator
performs best.
D. Ziou and R. Mohr summarised a SED (Selection of Edge
Detectors) system. This automatically selects edge detectors
and their scales to extract a given edge [2]. The system inputs
are location of an edge, the image, and set of constraints
related to the delocalization error and the computation time for
the desired quality. The results of the system are
characteristics of the given edge, the detectors, and their
scales. They used edge analysis, detector choice, and result
analysis for selecting edge detectors. D. Ziou and A. Koukam
summarised a SED system [3]. They used image structure
analysis involved segmentation of edge into a set of edgels,
detector choice, and result analysis.
Mike Heath et al. described a new experimental framework
for making quantitative comparisons using subjective ratings
made by people [4]. This approach is complement to signal-
based quantitative measures. They selected four edge detectors
Canny, Sobel, Nalwa-Binford, and Sarkar-Boyer for
comparison. They set three conditions edge detector,
parameter set, and image. They used ANOVA analysis.
Qiang Ji and Robert M. Haralick introduced a new criteria
for analytically evaluating different edge detectors without the
ground-truth information [5]. They adopted kernel-variance
criteria for comparing different edge detectors than the regular
convolution based. They studied performance of four edge
2
Proc. of the Intl. Conf. on Advances In Computer and Electronics Technology- ACET 2014.
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-024-8 doi: 10.15224/ 978-1-63248-024-8-01
detectors using synthetic test image. The experiment results
shows integrated edge detector outperforms the canny edge
detector at noise level 5. Further, they studied, performance
difference between the LOG zero-crossing perator and
Haralick’s facet zero-crossing operator. They confirmed that
both the techniques generate comparable results for kernel
sizes larger than 25 pixels. They also recommended not to use
kernel sizes less than 11 for LOG operators.
Laura Liu and David Zhang proposed a palm-line detection
approach to simultaneously extract structure and strength
features of palm lines by minimizing a local image area of
similar brightness to each individual pixel [6]. They tested the
proposed palm-line detection approach with canny edge
detector and susan edge finder on the public palmprint
database built by the Biometric Research Centre at the Hong
Kong Polytechnic University. The EER of the palm-line
detector is 1.0% which is the lowest one compared with the
Susan edge finder 2.3% and the canny edge detector 5.4%.
Pablo Hennings et al., designed multiple correlation filters
in sub regions of the palmprint [7]. They proposed a
segmentation stage that selects palmprint sub regions to train
the filters in a class-by-class basis using different edge-
detection operators. They used phase symmetry approach to
extract amplitude and phase measures of the signal at
particular scale and space locations. They computed difference
of the absolute values of the even and odd filter sequences at
each scale. Then, they computed a weighted average of the
difference images at each scale. They used PolyU database for
experiments. The experiment results were shown in figure 4
and 5. By mistake they mentioned figure 5 based on edginess.
The average EER was 0.0012% for using edginess and
0.0003% for using phase-symmetry edge detector, which was
claimed as better.
Rodrigo Moreno et al., defined a methodology for
evaluating edge detectors through measurements on edginess
maps instead of on binary edge maps [8]. These measurements
avoided possible bias introduced by the application dependent
process of generating binary edge maps from edginess maps.
The features of completeness, discriminability, precision and
robustness, were introduced. The R, DS, P and FAR-
measurements in addition to PSNR applied to the edginess
maps were defined to assess the performance of edge
detection. Well-known and state-of-the-art edge detectors had
been compared by means of the new proposed metrics. Results
had shown that it is difficult for an edge detector to comply
with all the proposed features.
Weiqi Yuan et al. proposed a palmprint principal lines
detection method [9]. This method employed priori knowledge
of statistical properties about palm lines. They considered
particular direction of principal lines based on the feature of
their valley type edges and minimum gray value. Further they
devised linking algorithm for broken lines. This scheme
avoided blind searching and enhanced the robustness. An
extraction rate (ER) index was defined to evaluate the effect of
the approach. They achieved 86.67% extraction rate.
C. Saravanan proposed an enhancement scheme for
palmprint using median filters for biometric applications [12].
The experiment results shows that the enhanced palmprint has
bright ridges compared to normal ridges identified palmprint.
The edge detection of palmprint is the most important
phase in the palmprint matching. Thus, a performance analysis
of edge detectors is performed and the results are discussed in
the following section.
III. Experimental Results
The palmprint line identification is an important phase in
the palm print matching. It is proposed to analyse the six well
known filters Sobel, Prewitt, Roberts, Log, Zero Cross and
Canny for identifying the palmprint lines. The figure 1 shows
one of the experiment image used for this experiment. The
figure 2 shows the experiment results of various filters and its
line detection.
Figure 1. Segmented Palmprint
Sobel
Prewitt
Roberts
Log
3
Proc. of the Intl. Conf. on Advances In Computer and Electronics Technology- ACET 2014.
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-024-8 doi: 10.15224/ 978-1-63248-024-8-01
Zero-cross
Canny
Figure 2. Result image of various edge detection algorithms
IV. Conclusion
From the experiment results shown in the figure 2, the
order of edge detectors are listed in terms of performance, 1)
Canny, 2) Sobel, 3) Prewitt, 4) Roberts, 5) Log and 6) Zero-
cross. The Canny and Sobel edge detectors are highly
recommended for palm print matching applications.
References
[1] L.J. Spreeuwers and F. van der Heijden, Evaluation of Edge Detectors
Using Average Risk, IEEE, 1992, pp. 771-774.
[2] D. Ziou and R. Mohr, An Experience on Automatic Selection of the
Edge Detectors, IEEE, 1992, pp. 586-589.
[3] D. Ziou and A. Koukam, The Selection of Edge Detectors Using Local
Image Structure, IEEE, pp. 366-370, 1995.
[4] Mike Heath, Sudeep Sarkar, Thomas Sanocki, and Kevin Bowyer,
Comparison of Edge Detectors: A Methodology and Initial Study, IEEE,
1996, pp. 143-148.
[5] Qiang Ji and Robert M. Haralick, Quantitative Evaluation of Edge
Detectors Using the Minimum Kernel Variance Criterion, IEEE, 1999,
pp. 705-709.
[6] Laura Liu and David Zhang, Palm-Line Detection, IEEE, 2005.
[7] Pablo Hennings, Marios Savvides, and B.V.K. Vijaya Kumar, Palmprint
Recognition with Multiple Correlation Filters Using Edge Detection, for
Class-Specific Segmentation, IEEE, 2007, pp. 214-219.
[8] Rodrigo Moreno, Domenec Puig, Carme Juli`a, and Miguel Angel
Garcia, A New Methodology for Evaluation of Edge Detectors, ICIP
2009, pp. 2157-2160.
[9] Weiqi Yuan, Sen Lin, Haibin Tong, Shudong Liu, and Sen Lin, A
Detection Method of Palmprint Principal Lines Based on Local
Minimum Gray Value and Line Following, IEEE, 2011.
[10] G. Padmavathi, P. Subashini, and P. K. Lavanya, Performance
evaluation of the various edge detectors and filters for the noisy IR
images, Sensors, Signals, Visualization, Imaging, Simulation and
Materials, pp. 199-203.
[11] Haruki Ota, Ryu Watanabe, Koichi Ito, Toshiaki Tanaka, and Takafumi
Aoki, Implementation of Remote System Using Touchless Palmprint
Recognition Algorithm, MoMM2010 Proceedings, ACM, 2010.
[12] Chandran Saravanan, Enhancement of Palmprint using Median Filter for
Biometrics Application, International Conference on Advances in
Science and Technology, (ICAST2014), Science Publications, pp. 16-18,
2014.
About Author (s): Chandran Saravanan is born in
Tiruchirappalli, Tamilnadu, India, on
01-01-1973. He has completed Ph.D.
from Department of Computer
Applications, National Institute of
Technology, Tiruchirappalli,
Tamilnadu, India, entitled “Analysis
and Modelling of Grey-Scale Image
Compression” in the year 2009.
He worked as Programming
Assistant at Bharathidasan University, Tiruchirappalli,
Tamilnadu, India, from 1996 to 2000. He worked as Computer
Programmer at National Institute of Technology,
Tiruchirappalli, Tamilnadu, India from 2000 to 2007. He is
working as Assistant Professor at the National Institute of
Technology, Durgapur, West Bengal, India, from 2007 to till
date. He has published 17 papers in International peer
reviewed Journals, 11 papers in national / international
conference, one book and two book chapters. Under his
supervision one student awarded Ph.D. degree in the year
2013 and one student submitted Ph.D. thesis and one student
doing Ph.D.
Dr. C. Saravanan is member of IEEE, Professional member of
ACM, Life member of CSI and ISTE, Senior member of
IACSIT, Singapore, member of IAENG, Hongkong. He is also
serving as Editorial Board Member and Reviewer for several
peer reviewed international journals
