Multi-feature approach for writer-independent offline signature verification

Abstract

Some of the fundamental problems facing handwritten signature verification are the large number of users, the large number of features, the limited number of reference signatures for training, the high intra-personal variability of the signatures and the unavailability of forgeries as counterexamples. This research first presents a survey of offline signature verification techniques, focusing on the feature extraction and verification strategies. The goal is to present the most important advances, as well as the current challenges in this field. Of particular interest are the techniques that allow for designing a signature verification system based on a limited amount of data. Next is presented a novel offline signature verification system based on multiple feature extraction techniques, dichotomy transformation and boosting feature selection. Using multiple feature extraction techniques increases the diversity of information extracted from the signature, thereby producing features that mitigate intra-personal variability, while dichotomy transformation ensures writer-independent classification, thus relieving the verification system from the burden of a large number of users. Finally, using boosting feature selection allows for a low cost writer-independent verification system that selects features while learning. As such, the proposed system provides a practical framework to explore and learn from problems with numerous potential features. Comparison of simulation results from systems found in literature confirms the viability of the proposed system, even when only a single reference signature is available. The proposed system provides an efficient solution to a wide range problems (eg. biometric authentication) with limited training samples, new training samples emerging during operations, numerous classes, and few or no counterexamples.

Date	20 Oct 2010
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Robert Sabourin (Supervisor) & Éric Granger (Co-supervisor)