الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 241 |  |  |
| | | from | 241 | | |
Abstract
Automatic signature verification has gained an intense research area compared to other biometrics because of the widespread use of signatures as a personal authentication method as well as the high level of acceptance of signatures from the legal and societal view. Signature verification systems divided into two main categories: offline and online. Offline signature verification processes only the signature’s image captured by a scanner (or camera). It can be useful in verifying signatures found on bank checks or documents. Online signature verification processes the data captured by pressure-sensitive tablets that samples the hand-drawn signature at regular time intervals. Online signature verification approach is favorable because in order to produce a forgery, an impostor has to reproduce not only the static image of the signature but also the gesture of signing which is more difficult to imitate than the image of the signature. Signature forgeries are divided into four types: random forgery, simple forgery, skilled forgery and self-forgery. The first type is called random forgery, in which the forger will try to log in to the system with a not-real claimed identity using any scribbled written signature. The second type is called simple forgery or over-the-shoulder forgery, in which forgeries are captured after the forger watches the genuine signature is being written in front of him. The third type is called skilled or professional forgery, in which the forger will have unrestricted access to genuine signature samples and has a complete knowledge about the personal gesture of signing and the image of the signature. The fourth type is called self-forgery, in which the original signer tries to login with his identity but with a not-real signature. Self-forgery is produced by the original signer to authorize a transaction with a priori intent of later denying this authorization by pointing to the visual discrepancy between the genuine signature and the signed one.
The objective of this thesis is to develop a multi-modal signature verification system that can be implemented on the real world to authenticate signatures acquired using a digitizing tablet and to prevent all these types of forgeries. Our online signature verification system consists of three verifiers: the first verifier is a local-based verifier, the second verifier is a global based verifier and the third verifier is a shape-based neural network verifier. The local verifier represents the signature by nine time-based functions and transforms each function to the wavelet domain. Each transformed function in the test signature is aligned with the corresponding transformed function in the reference set using extreme point warping technique. Finally, one-way-ANOVA is used to analyze the variance between the warped functions. The 18 local similarity scores (9 detail coefficients signals + 9 approximation coefficients signals) are combined together by using the average of them and this average is used as an output of the local verifier. The second verifier uses Mahalanobis distance to calculate the similarity between the feature vector of the test signature and the reference set of the claimed identity. The third verifier uses a multi-layer perceptron neural network to recognize the shape of the signature. The x and y signals are used to generate 2D signature image and is then recognized by the neural network. The output of the neural network is a similarity measure that measures to what extent is the image of the test signature is similar to the claimed identity’s signature image. The three scores are combined using harmonic mean and then this mean is compared to a threshold to accept or reject the signature. Several experiments were made to evaluate the effectiveness of the proposed technique. The proposed system was tested on two databases: first is our developed database and second is the Session II of the Signature Verification Competition (SVC’2004) database. Experimental results are very encouraging and we obtained the lowest values for false rejection rate and false acceptance rate compared to other techniques in the state of the art of online signature verification.