International Journal of Computer Application Issue 2, Volume 3 (june 2012) Issn: 2250-1797

Hand-written signature is widely used for authentication and identification of individual. Since the hand-written signature can be random, because of presence of various curves and features, techniques like character recognition cannot be applied for signature verification. In this paper we present an off-line signature verification system based on combination of features extracted such as global feature, mask feature and grid feature. In this database of signature taken from 5 different persons,11 genuine and 11 forged samples from each person, and four feature are extracted for all the samples namely, centroid ,trisurface, six-fold surface and shape number feature . To verify the similarity between our template and claimed signature, we use the Euclidean distance in the feature space and based on a specific threshold, the signature is declared original or forgery. The threshold used in the proposed technique can be dynamically changed according to the target application. Then we calculate FRR and FAR rate to measure the performance. Keywords— Signature verification, Feature extraction, Euclidean distance, Skilled Forgery, False Acceptance Rate (FAR), False Rejection Rate(FRR). ______________________________________________________________________________ INTRODUCTION Signature verification is a biometric verification which is an important research area targeted at automatic identity verification applications such as legal, banking and other high security environments. Such applications need their own exclusive software for signature verification. Biometrics based authentication systems are better in terms of security than traditional authentication techniques such as passwords etc. It is due to the fact that biometric characteristics of every person are unique and cannot be lost, stolen or broken. There are two types of biometrics: Behavioural and Physiological. Handwriting, speech etc. come under behavioral biometrics. Iris pattern, fingerprint etc. are part of physiological biometrics. There are two methods for signature verification: Offline and Online, which depends on the signature acquisition method. In offline signature verification, after having complete signature on the paper, it can be acquired from scanners or cameras. In online method, during signing process, it can be acquired in parallel with digitizing tablets or any other special hardware. The purpose [7] of signature verification is to classify the input signature as genuine or forge by matching it against the database signature image using some distance INTERNATIONAL JOURNAL OF COMPUTER APPLICATION ISSUE 2, VOLUME 3 (JUNE 2012) ISSN: 2250-1797 Page 347 measure. Forgery means that an individual is trying to make false signatures of any other individual to become authenticated. There are three types of forgeries: (1) Random Forgery: This is also known as simple forgery and is very easy to detect. The signer creates a signature in his own style by just knowing the name of an individual whose sign is to be made. (2) Unskilled Forgery: The signer creates a signature after observing the signature once or twice without any prior experience. (3) Skilled Forgery: The signer may be a professional in copying signatures. He creates a signature after having a good practice over it. Such signatures are most difficult to detect. A lot of research has been conducted on offline signature verification. In [2] signature verification is based on quasi multi-resolution technique using Gradient, Structural and Concavity (GSC) features. In the signature verification is based on Feature Point Extraction method. Verification is performed by comparing the already trained feature points with the feature points extracted for test image. In each pixel of the signature has been studied and extracted the end points of the signature. Verification is based on structural features such as perimeter, area, circulatory measure, rectangularity measure, minimum enclosing rectangle and form factor is a verification technique using global and local feature extraction in high pressure regions of an image. In this paper, all experiments have been performed on skilled forgeries. The paper includes signature acquisition, signature preprocessing, feature extraction and verification techniques. METHODOLOGY In order for us to be able to perform verification of signatures, there are several steps that need to be completed first. The first step is pre-processing and in this step the image is converted from a bmp and into a binary image which makes the feature extraction simpler. Once the image is binarised, we begin our feature extraction phase of the signature. These features are combined to form our template for comparison with test signatures. Verification is then performed using the Euclidean distance to obtain a measure of similarity between the claimed signature and the template signature. Figure 1[1] illustrates the entire process. Figure 1. Procedure to identify/verify a signature from the template INTERNATIONAL JOURNAL OF COMPUTER APPLICATION ISSUE 2, VOLUME 3 (JUNE 2012) ISSN: 2250-1797 Page 348 A. Feature Extraction The choice of a powerful set of features is crucial in signature verification systems. The features that are extracted from this phase are the inputs in the template creation phase. We use a feature vector to uniquely characterize a candidate signature. These features are geometrical features, which mean they are based on the shape and dimensions of the signature image. These features are centroid, trisurface, Six-fold surface and shape number feature and the description of all are as follows: 1) The centroid feature: This feature is related to the angle of the signatures pixel distribution, the „centroid‟[1]. The first step in finding the centroid consists of splitting the image into two equal parts. Then in each part of the image, the centre of gravity was calculated (A and B in Figure 2). The angle between the horizontal axis and the line formed by joining the two centers of gravity was the feature that was added. This angle is indicated in yellow in Figure 2. Equation (1) is followed to calculate the angle in question. Equation (2) is followed to make sure the angle is represented by a value between 0 and 1. Figure 2 The Centroid Feature α= arcsin (height/hypotenuse) / π (1) Centroid= α+0.5 (2) 2) The Trisurface feature: The surface area of two visually different signatures could be the same. For the purpose of increasing the accuracy of feature describing the surface area of a signature, the „Trisurface‟ feature was investigated, as an extension, in which the signature was separated into three equal parts, vertically. The surface area feature is the surface covered by the signature, including the holes contained in it. The total number of pixels in the surface was counted, and the proportion of the signature‟s surface over the total surface of the image was calculated. This process was used for the three equal parts of the signature, giving three values between 0 and 1. Figure 3 illustrate this concept. Figure 4 The trisurface area INTERNATIONAL JOURNAL OF COMPUTER APPLICATION ISSUE 2, VOLUME 3 (JUNE 2012) ISSN: 2250-1797 Page 349 3) The sixfold-surface feature: The concept with this feature is similar to that of the trisurface feature with two main differences. The first difference is that this feature yields six features. The second difference is that centers of gravity are determined to assist in the calculation of the sixfold surface features. The first step is to divide the image into three equal parts, the same way as the trisurface feature is done (Figure 3). In each of the three areas, we must calculate the bounding box for that area. In each bounding box, the centre of gravity is then calculated. For each bounding box, the area of pixels above and below the centre of gravity is calculated, thus yielding you with six features. Each area is represented as a percentage of the entire image area. Figure 4 illustrates this concept. Figure 4. The sixfold-surface feature 4) The shape number feature: The shape number of a boundary, generally based on 8-directional Freeman chain codes, is defined as the first difference of smallest magnitude. In this feature first we compute the chain code of a boundary depends on the starting point, however the code can be normalized with respect to the starting point by treating it as a circular sequence of direction numbers and redefining the starting point so the resulting sequence of the numbers forms an integer of minimum magnitude and then we normalized for rotation in increments of 45 o by using the first difference of the chain code instead of the code itself . This difference is obtained by counting the number of direction changes. After computing the chain code difference then reorder this to create the minimum integer, this is called as shape number. Figure 5 illustrate signature sample, then normalized the signature and filter that to remove background noise and then compute the boundary image (Figure 6) and subsample the image and after that connected sequence of a image (Figure 7) is generated then compute the chain code, difference of chain code and shape number. Figure 5 Signature sample. INTERNATIONAL JOURNAL OF COMPUTER APPLICATION ISSUE 2, VOLUME 3 (JUNE 2012) ISSN: 2250-1797 Page 350 Figure 6 Boundary Image. Figure 7 Connected sequence of a image. B. Signature Verification The verification of the signatures is accomplished by using the Euclidean distance. The Euclidean distance is defined as the “ordinary” distance between two points that one would measure with a ruler. Once a claimed signature is entered into the system, the Euclidean distance between the claimed signature and the signature template is calculated. Depending on the threshold of the system, the signature will either be correctly identified as genuine or identified as a forgery. The Euclidean distance between points P = (p1, p2, ..., pn) and Q = (q1, q2, ..., qn) is defined as: RESULTS AND DISCUSSION For training and testing of the system many signatures are used. To test the system, a subset of this database was taken comprising of 11 genuine samples and 11 forged samples taken from each of 5 different persons. The features of template signature are computed and stored in an array, then the claimed signature is entered into the system, it is compared against the template vector using Euclidean distance and average all the difference features and based on the certain threshold value (in Table 1), the Signature verification system can produce two type of errors: INTERNATIONAL JOURNAL OF COMPUTER APPLICATION ISSUE 2, VOLUME 3 (JUNE 2012) ISSN: 2250-1797 Page 351 FAR (False Acceptance Rate): The false acceptance ratio is given by the number of fake signatures accepted by the system with respect to the total number of comparisons made and is given by: FAR = No. of forgeries accepted *100 No. of forgeries tested FRR (False Rejection Rate): The false rejection ratio is the total number of genuine signatures rejected by the system with respect to the total number of comparisons made and is given by: FRR = No. of genuines rejected *100 No. of genuines tested FAR and FRR are the two parameters used for measuring the performance of any signature verification method .The purpose of verification is to reduce FAR and FRR. And the signature verification results shown in (Table 1). And Figure 8 shows the graph FAR versus FRR against threshold. Table 1. Signature verification Results Threshold FAR(%) FRR(%) 25 72 18 30 54 18 35 45 18 40 45 27 45 45 36 50 36 45 55 36 45 60 36 54 65 18 72 Average 43.6 38.1 Figure 8 FAR versus FRR against threshold value. INTERNATIONAL JOURNAL OF COMPUTER APPLICATION ISSUE 2, VOLUME 3 (JUNE 2012) ISSN: 2250-1797