Automatic Bank Check Processing and Authentication using Signature Verification

Madasu, Vamsi-Krishna (2006). Automatic Bank Check Processing and Authentication using Signature Verification PhD Thesis, School of Information Technology and Electrical Engineering , University of Queensland.

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
n01front_Madasu.pdf n01front_Madasu.pdf application/pdf 246.99KB 2
n02content_Madasu.pdf n02content_Madasu.pdf application/pdf 5.20MB 1
Author Madasu, Vamsi-Krishna
Thesis Title Automatic Bank Check Processing and Authentication using Signature Verification
School, Centre or Institute School of Information Technology and Electrical Engineering
Institution University of Queensland
Publication date 2006
Thesis type PhD Thesis
Supervisor Professor Kurt Kubik
Abstract/Summary Automatic processing of bank checks is a challenging research topic in the field of document analysis and recognition. Bank check processing is the process of automatic segmentation and recognition of the different data fields present on the bank check. In this process, the user entered handwritten information is automatically extracted from the check and recognized by a computer. Authentication is performed by comparing the handwritten signatures with the reference samples provided by the user. For the purpose of check processing, we have proposed a generic method for the automatic segmentation and identification of the information fields on a bank check. The uniqueness of this approach lies in the fact that it doesn’t necessitate any prior information about the layout of the check and requires minimum human intervention. The various fields are segmented on the basis of their connectivity. Recognition of the fields is achieved based on the four newly devised fuzzy features, namely, average response, entropy, energy and quotient response. In addition to the proposed fuzzy features, Zernike moments have also been explored. A simple fuzzy logic based recognition approach is used to identify the various fields. The fuzzy features are able to identify all fields with reasonable accuracy but Zernike moments are found to distinguish the written matter from the printed matter only; hence these are used for coarse classification only. Automatic verification of handwritten signatures is fundamental to the authentication of bank checks. This thesis presents two approaches for handwritten signature verification using two different sets of features. In the first approach, additive fuzzy modeling has been employed to track the intrinsic variations in signatures samples of an individual for the twin purposes of signature verification and forgery detection. Innovative normalized angle features that uniquely characterize a signature are extracted by constructing a special grid which encloses the signature and divides it into ninety six local boxes. These features are then fuzzified by an exponential membership function, which has been modified to include two structural parameters. The structural parameters are devised to deal with innumerable variations in handwriting styles and personal characteristics. The membership functions constitute weights in the Takagi-Sugeno (TS) fuzzy model. The optimization of the output of the TS model with respect to the structural parameters yields their estimated values. Two cases are considered. In the first case, the coefficients of the consequent part of the rule are fixed so as to yield a simple form of TS model and in the second case, the coefficients are adapted. In this formulation, each fuzzy rule is constituted by a single feature and is implemented on the signature database. In the second formulation, only one rule encompassing all the features is considered. Experimental results demonstrate that the simple form of TS model in the first formulation is better than the one with coefficients adapted. In the second approach, edge features based on direction and hinge distribution are extracted from the signatures to create the knowledge base. Signature recognition is then performed using the fuzzy model derived from the Choquet integral. The output function in this model combines the fuzzy measures from input fuzzy sets and the resulting system is called non-additive fuzzy system unlike the output function in TS model where each coefficient corresponds to its own fuzzy set. This system is meant for modelling of the input fuzzy sets that have overlapping information. However, the performance of this system is found to be inferior to that of the additive fuzzy system as it is not able to detect forgeries effectively. To improve its performance, the decision from its model is fused with the decision of the simple TS model and a neural network classifier. Experimental results illustrate that although the fusion of verification algorithms produces better performance than any of the fused methods individually, it is still worse than the first approach using the grid features. Finally, a comparative analysis of well known static signature verification schemes is made on our signature database. The results clearly demonstrate the efficacy of the proposed methods.

Citation counts: Google Scholar Search Google Scholar
Created: Fri, 21 Nov 2008, 14:50:07 EST