28-11-2012, 04:18 PM
The Ultimate Signature Identifier
The Ultimate Signature Identifier.pdf (Size: 360.69 KB / Downloads: 30)
Abstract
Biometric identification is the need of hour, as
automatic recognition systems are such biometric techniques to
look upon. Accurate automatic recognition systems are
important for a wide range of applications such as banks,
restricted areas, government classified areas etc. As traditional
identity recognition methods such as pins, passwords etc suffer
from some fattle flaws and are unable to satisfy the security
requirements. The paper aims to consider a more reliable
biometric feature, signature verification for the considering.
The paper presents an experimental comparison of different
signature verification methods. It can be effectively employed
to reduce the risk of forgery by releasing the trouble of
carrying ATM cards by the users, by employing a signature
verification system. The paper compares four of the most
extremely efficient methods used for signature verification like
CDTW (Continuous Dynamic Time Warping), DTW (Dynamic
Time Wrapping), Vector Quantization followed by HMM
(Hidden Markov Model) employed for signature verification
and the best method was stated. The paper also aims at
improving the efficiency by integrating the methodologies
introduced in the paper with each other. The methods were
tested on synthetic and further applied on real time signature
datasets. It is proved that superiority is achieved by
combination of different methods.
INTRODUCTION
Biometrics is the need of the hour as it is heavily required
for security in many critical scenario’s like banks, ATM’s,
credit card transactions etc. Recent advancements in the field
of biometrics have outreached human’s performance in
knowing an individual’s identity. Biometrics is mainly
divided into two groups: physiological and behavioral
biometrics. Physiological traits are related to the shape of the
body like face, finger print etc while behavioral are related to
the behavior of the person. Many methods have been applied
for physiological model like speech recognition, face
detection but major flaws have been observed in the
methodologies along with a lot of computational cost.
In today’s biometrics system and in this modern world of
sophistication, a person finds it difficult to take care of
plastic cards, pin codes etc which may get damaged in one or
the other way can be forgotten. So the aim of this paper is to
search for an ultimate identifier which identifies a person
without the need of these prerequisites mentioned above.
In order to overcome these disadvantages, signature
verification as behavioral models has been used.
Signatures are composed of special characters and flourishes
therefore most of the time they can be unreadable. Also
intrapersonal variations and interpersonal differences make it
necessary to analyze them as complete images and not as
letters or words put together [1]. As signatures are the
primary mechanism both for authentication and authorization
in legal transactions, the need for research in efficient automated
solutions for signature recognition and verification has
increased in recent years.
Signature verification can be further classified into online
and offline signature verification. In online signature
verification, a computer tablet with a pressure pen, stores the
dynamic information which can be further used for
identification while offline signature verification deals with
static image of signature which is scanned and can be further
used for verification.
RELATED WORK
In order to achieve successful signature verification, the
problem of combining disparate forms of signature
information, including both continuous and discrete features,
parametric and signal-based methods for signature
representation [3] must be addressed [4]. The combination of
different sources of information about a signature, in the
form of different feature sets and classification methods,
provides an opportunity to develop an improved level of
verification compared to the use of a single set of
descriptors[5]. Online signature verification methods vary in
classification methods, preprocessing and feature selection.
A number of classification methods have been discussed in
[2]. The methods focus on differentiation of sine and cosine
angle values with respect to consecutive points in x and y
plane followed by grey values in adjoining pixels. Focusing
on the classification methodology, different approaches can
be found to measure the similarity between test signature and
signer model. Dynamic time warping [2] and hidden Markov
model (HMM) [3] are widely used for signature verification.
Vector quantization (VQ) [4] and pattern recognition
algorithm has also been tested [4] in this field. Dynamic time
warping [1] and hidden Markov model(HMM)[3] are two of
the most widely used approaches followed by Vector
quantization(VQ) In the Signature Verification Competition
2004 (SVC04), DTW and HMM based systems were shown
to be the most competitive algorithms.
CONCLUSION
The results obtained from real time simulation by
CDTW, DTW, HMM, VQ+DTW and VQ+CDTW applied
on real time data are presented in the paper. The following
presents a concise summary with the conclusions related to
the performance of algorithms. VQ+CDTW performs much
better than all the algorithms presented in the paper, with the
highest accuracy. The reason for the high accuracy lies in the
integration of two highly efficient methodologies. CDTW
gives minimum distance between two curves to a much
higher precision as compared to DTW while it performs
much better than HMM. The value given by CDTW is higher
than DTW and HMM signifying that it can measure to much
higher accuracy, leading to a very less false acceptation ratio
as compared to the other method. In the process of CDTW,
along with the points used in DTW process, some more
intermediate points are introduced making the distance
between the curves function somewhat more continuous.
Theses intermediate points an extra edge over the DTW and
HMM method. The CDTW distance calculated between the
curves helps in analyzing the originality of data. CDTW is
observed to be more accurate as compared to DTW as DTW
works only on end points while CDTW computes the
similarity on end as well as intermediate points. The distance
matrix clearly demonstrates the capabilities of the system to
differentiate forge signatures effectively from the original
signature by a large margin. On the other hand, vector
quantization has proved to be an extremely effective
methodology due to its high recognition rate as compared to
other methodologies.