Abstract—In Then each features will be ranked based on

Abstract—In this paper we collect 1210
global features of dynamics signature. All features are collected from previous
research and we also add global features from FRESH framework.  The purpose
is to select relevant features from those features set. For doing that, we
compute the importance score of each features using two methods: Information Gain Ratio and Correlation. Then each features will be
ranked based on its score. After that, we divide 1210 features into 10, 20, 30,
and so on until all 1210 features set are used in ascending order of the rank. Then,
each features set is validated with Random Forrest, SVM, and Naïve Bayes classifier
using 10 folds cross validation. The experiment was conducted in SVC2004
dataset task 1. The result show that, 120 first features from correlation
method is the most relevant features. And therefore, those 120 global features
is a good candidate for dynamic signature verification and will be used for the
next research using dataset collected from mobile device.

Signature, Global Features, Features Selection, Information Gain Ratio,

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now

I.     Introduction

Signature are
non-invasive and people are familiar with the use of signature in everyday life
1. Signature have long been used as a way of
verifying a person’s identity. In general, signature are used as a legal means
of verifying document ownership in financial and administrative institutions.
These documents are often used as facts or legal evidence, for example, can be
used as evidence of who owns certain property and is used as proof of the
amount of salary received in the employment contract. Therefore, there have
been many studies to establish signature verification system in digital documents. Based on the data
acquisition method, the signature verification system can be categorized into
two groups: static or offline signature verification and dynamic or online
signature verification. A static signature is a signature that the acquisition
process is done after the signature process is complete with acquisition tools
like scanner. In this case, the signature is represented as an image with gray
level image {S(x,y)}0?x?X, 0?y?Y, where S (x, y) denotes the gray level at
position x and y in the image. While a dynamic signature is a signature that the
taking process use an acquisition device like digitizer or tablet that
generates an electronic signal during the signing process. In this case, the
signature is represented as the time function sequence{S(n)}
n=0,1,…,N, where S
(n) is the signal value taken at time n?t from the signature
retrieval process, where ?t is the period sampling. Such signals are
also commonly referred as time series. Since dynamic signature consider
temporal information, then verification of dynamic signature is more accurate
than static signature about 5-25% 2.

      Dynamic signature features are
broadly divided into two groups: functional features and global features 3. A functional feature is a dynamic feature that
uses all the points of the time function for the signature verification
process. For example, the function x(n) is a function of the x-coordinate
position, then its functional feature is the overall point of x-coordinate from
the beginning of signing time up to the end of the signing time (n = 1 to n =
N). Unlike functional features, global features do not use the entire point for
the verification process, but only use some values ??that represent the entire
point. In general, dynamic verification with functional features shows better
performance when compared to global features, but on the other hand functional
features require greater computing and storage resources. In addition, there is
a need to reduce the size of signature data in commercial applications,
especially for large-scale systems and systems with sources limited such as
smart cards 3. Therefore, there is a high interest to find an
approach whereby a person’s signature can be solidly represented to reduce
storage and computacy requirements. Global features match those needs. There
are two aspects of global features of dynamic signature, they include shape
aspect and dynamic aspect that play a complementary role in distinguishing
genuine or fake signature. For example, when a forger attempts to falsify every detail of a signature,
the signature results tend to be less suitable in terms of its dynamic
features, and vice versa.

Figure 1 is a
research roadmap of dynamic signature verification. Worfklow is divided into
two parts. The first part is what will be done in this paper. The goal is to
collect global features from previous research and then rank them by their importance
score. The In this paper we will use two importance measurement algorithm, they
are Information Gain Ratio and Correlation. By using the importance score we
will rank the global features from the most important features to the less
important features. After that, we will select features subset that give the best
performance. To test the performance of features subset, we will use ten folds
cross validation using three classifier, they are Random Forrest, SVM, and
Naïve Bayes. The final result of this research is the the most relevant
features set. Those features set will be used in the next research on
dynamic signature verification using dataset collected from mobile device. For
this experiment, SVC2004 dataset task 1 will be used because the data provided
in that dataset is similar with the data provided in mobile device. See 4 for
the detail.