Writer-independent offline handwritten signature verification system based on the dichotomy transformation, prototype selection and feature selection

Abstract

High number of writers, small number of training samples per writer with high intraclass variability and heavily imbalanced class distributions are among the challenges and difficulties of the offline Handwritten Signature Verification (HSV) problem. A good alternative to tackle these issues is to use a writer-independent (WI) framework. In WI systems, a single model is trained to perform signature verification for all writers from a dissimilarity space generated by the dichotomy transformation. Among the advantages of this framework is its scalability to deal with some of these challenges and its ease in managing new writers, and hence of being used in a transfer learning context. In this work, a deep analysis of this approach is presented, highlighting how it handles the challenges as well as the dynamic selection of reference signatures through fusion function, and its application for transfer learning. All the analyses are carried out using the instance hardness (IH) measure. By having these findings at the instance level, we develop an approach that uses prototype selection (Condensed Nearest Neighbors) and feature selection (based on Binary Particle Swarm Optimization) techniques well suited to our WI-HSV scenario. These techniques allowed us to handle the redundancy of information in both sample and the feature levels present in the dissimilarity space. Specifically in the feature selection scenario, we also propose a global validation strategy with an external archive to control overfitting during the search process. The experimental results reported herein show that the use of prototype selection and feature selection in the dissimilarity space allows a reduction in its redundant information and the complexity of the classifier without degrading its generalization performance. In addition, the results show that the WI classifier is scalable enough to be used in a transfer learning approach, with a resulting performance comparable to that of a classifier trained and tested in the same dataset. Finally, using the IH analysis, we were able to characterize “good” and “bad” quality skilled forgeries as well as the frontier region between positive and negative samples.