Dynamic Signature Verification Using Pattern Recognition



With the growth of web enabled services and e-commerce, tremendous amount of information is now readily available on the Internet. A large proportion of this is classified information, which has to be protected against unauthorized access. Password or PIN can be used in conjunction with digital signature, for verification of the identity of users. This paper proposes a dynamic handwritten signature verification based access control system that can be employed in the legal, banking and commercial domains for designing secure information retrieval systems. The dynamic handwritten signature in this system is captured by using a digital tablet or PDA (Personal Digital Assistant) with contact sensitive acquisition system. After preprocessing, the signature data is compared with the templates of authorized signatures by employing an innovative neuro-fuzzy pattern recognition system based on sensing the pressure variable and total time required for executing the signature for uniquely identifying the potential user. The error in matching is used to arrive at the decision regarding permission or denial of access to the classified document. The neuro-fuzzy technique applied in the dynamic signature system is based on evolving fuzzy neural network. This technique has been tested on signatures drawn from signature verification competition database obtained from the internet. Encouraging results show that this technique is a good candidate for the development of practical applications.





1.1       Background of the study

Signatures are composed of special character and flourishes and therefore most of the time they can be unreadable. Also, intrapersonal variations and interpersonal differences make it necessary to analyze them as complete images but not as letters and words put together. Signatures have been the primary mechanism both for authentication and authorization in legal documentation in recent years.

Based on different applications, signature verification system can be operated in two different modes (Plamondon and Srihari, 2000, Seiler et al., 1996); online and offline mode. In the online mode, the signature verification is dealing with the instant inputs from the system such as credit card verifier. For offline mode, the verification is done on the recorded signatures such as bank’s document verification (Dimauro et al., 1997,


Generally, signature verification system can be categorised into two types: dynamic and static. The dynamic signature verification system is dealing with signal processing while the static signature verification system is more on image processing. Some techniques applied in static signature verification systems are neural networks (Bajaj and Chaudhury, 1997; Huang and Yan, 1997; Karouni et al., 2011), model based approaches (Huang and Yan, 2002; Wen et al., 2009) and wavelets transform (Deng et al., 1997) . Meanwhile, Dynamic Time Warping (DTW) (Fenton et al., 2006) and Gaussian Mixture Modelling (GMM) methods (Miguel-Hurtado et al., 2007, 2008) have been introduced for dynamic automated signature verification system. Basically, DTW is used in pre-processing to remove the intrinsic variability from user signature by aligning the acquired signal. GMM is used to model the probabilistic distribution of the set of pseudo-distances and to calculate the likelihood ratio between the sample and reference signature.


There are other approaches which based on the concept of filters (Tanaka and Bargiela, 2005). Firstly, global features of the signature, such as average velocity are considered through Euclidian distance. In the second filter, local features are considered. Strokes are segmented using the minima of the velocity and encoded before comparing them using DTW (Miguel-Hurtado et al., 2007, 2008) and signer -specific thresholds. On the other hand, Linear Prediction Coding (LPC) cestrum and Neural Networks (Wu et al., 1997) are proposed in the dynamic signature verification system. LPC is used in the pre-processing stage and its coefficients are used as the input to the neural networks. The neural networks (Mailah and Han, 2008) mostly used in the verification process. Besides that, performance could be improved by fusing static and dynamic signature verification techniques (Alonso-Fernandez et al., 2009).


Dynamic (or on-line) handwritten signature is one of the modalities within bio-metrics that has vital importance in terms of establishing the identity of an individual, mainly because of the social and legal acceptance of handwritten sig-natures as a means for person identi cation in the day-to-day life [1] . The latest innovations in touch screen technologies have provided a feasible environment for dynamic signature veri cation in smart phones and mobile scenarios. Despite the fact that the technology innovations have made it to a point where dynamic signature acquisitions is easy in smart phones, inherently signature veri cation faces some challenges which are in general applicable to either smart phones or more ergonomically designed signature pads. The latter sce-nario (i.e., pen-based digitizing tablets) is commonly studied in the signature veri cation literature [1] . The purpose here is to adapt established technology previously developed fordigitizing tablets for smart phones, and then evaluate its performance and discuss some of its particularities.

More specifically, two of the challenges faced in signature verification are intra-class variability where the individual has slight variations in their own signature writing styles over a period of time, and inter-class variability where some other person tries to mimic or simulate the signature of an individual to get an illicit access through a signature verification system. Traditionally, it has been thought that these sources of variability, specially the intra-class variabil-ity, is much higher in mobile scenarios compared to desktop digitizing tablets for signature veri cation, which results in degraded veri cation performance in mobile scenarios [2] . Nevertheless, this comparison has always been evidenced using limited mobile acquisition devices, far from the capabilities of state-of-the-art touch-based and stylus-based smart phones [3] .


With regard to the inter-class variability, forgeries can be classified as two types, random forgeries and skilled forgeries. In random forgeries the forger has no information regarding the target signature, whereas in case of skilled forgeries the forger has knowledge about the target signature [1] . In the present work, only random signatures are considered.


With respect to the kind of information used in the recognition process, the signature verification can be classified into feature based systems and function based systems [4] . In feature based systems, a set of global features derived from the signature sample is used, whereas in function based systems, temporal sequences which encapsulate the local properties of the signature samples are used.


As introduced before, in this paper we present the results of the adaptation of an already existing dynamic signature verification algorithm for smart phones. As with any technology, there are some pros and cons associated with smart phones in the context of signature verification, though there is a growing interest in the use of portable devices for personal authentication. For signature verification, one advantage is related to the acquisition hardware, as with touch or stylus based smart phones there is no need for specialized external hard-ware for signature recognition. Most smart phones come with enough computing power, good quality touch screens and supports pen based input which makes them a feasible platform for dynamic signature verification.


Coming to the challenges, usually smart phones do not provide big display areas user interaction and leads to large intra-class variability, the quality of the signature acquisitions can show high disparity based on the quality of the touch screens and the amount of information that can be captured is limited as pressure, pen-azimuth and other attributes which could lead to improved performance cannot be captured. Finally, in smart phones it is also important to consider the security of the templates [5] .


As also introduced before, the public domain evaluations of dynamic sig-nature verifications like BioSecure Multimodal Evaluation Campaign (BMEC-2007) [3] , and BioSecure Signature Evaluation Campaign (BSEC'2009) [2] have shown that the performance of dynamic signature verification with databases captured on handheld devices are significantly lower compared with databases captured on ergonomically designed signature pads or tablet PCs.

The main reasons for such performance variations are the challenges discussed previously, such as small display area, quality of touch screen sample acquisitions and limited attributes of the samples acquired.


  • Problem Statement

Signature verification techniques utilize many different characteristics of an individual’s signature in order to authenticate that individual ( Vacca,  2007). The advantages of using such an authentication technique are; (i) signatures are widely accepted by society as a form of verification ( Kung et al.,  2004), (ii) information required is not sensitive and (iii) forging of an individual’s signature does not mean a long-life loss of that individual’s identity. The general idea of this re-search is to investigate a signature verification technique which is not costly to built, user friendly in terms of configuration, robust against imposters and is reliable even if the individual is under different emotions.


In signature verification application, the signatures are processed to extract features that are later fed into a classifier. The task of the classifier is to assign the signature features to classes of individuals. The selection of signature features is critical in determining the performance of a signature verification system. In this research, the features were selected according to certain criterions. Mainly, the features have to be small enough to be stored in a smart card and does not require complex classification techniques.


There are two ways of validating a signature. They are static and dynamic. Static features are comprised of features which are extracted from signatures that are recorded as an image whereas dynamic features are extracted from signatures that are acquired in real-time ( Faundez-Zanuy,  2005;  Plamondon and Srihari,  2000). These feature types can be broken down into two types which are function based and parameter based features.


The function based features describes a signature in terms of a time-function. Examples of function based features include position, pressure and velocity (Di-mauro et al.,  2004). While the performance of such features is well known to researchers in accurately verifying signatures, they are not suitable in this case due to the complexity of its matching algorithm. Hence, the use of parameter based features is more appropriate.


Even though it is critical to select a suitable set of features to be extracted, emphasis has to be put into selecting an appropriate classifier for the features selected. Some classifiers do not work for certain type of features, for example Hidden Markov Model (HMM). The HMM classification technique for signature verification has been proposed by many researches such as Igarza et. al. (Igarza et al.,  2003) and Muramatsu et. al. ( Muramatsu and Matsumoto,  2003). The main issue in using HMM is modeling the extracted features in Markov Model. Moreover, the larger the amount of features, the more complex the HMM would be. There are many other classification techniques available that have been pro-posed by Srihari et. al. ( Srihari et al.,  2004), Rioja et. al. ( Rioja et al.,  2004) and Sakamoto et. al. ( Sakamoto et al.,  2001). These would be discussed in detail in chapter 2.

It is important to take into account external factors when investigating a signature verification technique. Signature verification applications are used in our daily lives and will be exposed to human emotions. The system has to be reliable in accurately verifying an individual’s signature even if he/she is under different emotions. Sackheim ( Sackheim,  1990), Gardner ( Gardner,  2002), Lange et. al ( Lange et al.,  2006) and Yank ( Yank,  1991) have shown that handwriting of a person is affected by their emotions. Most of the techniques which have been proposed by researchers have not been tested against people’s emotions.


  • Research Aim and Objectives

The general idea of this research is to investigate a signature verification technique which is not costly to build, user friendly in terms of configuration, robust against imposters and is reliable even if the individual is under different emotions. The main aim of this research is to:

  • Choose a suitable features required for a robust signature verification tech-nique, yet inexpensive to build and user friendly in terms of configuration.
  • Investigate the performance of selected classifiers which are suitable for classifying the chosen features.



Leave a Reply