On-line signature verification system with failure to enroll managing

TL;DR

This paper presents an online signature verification system that manages enrollment failures through an intelligent enrollment process, improving verification accuracy and identifying users who cannot be enrolled, using a DCT-based feature extractor.

Contribution

The work introduces a novel intelligent enrollment method for automatic rejection of low-quality samples, enhancing verification accuracy and managing enrollment failures.

Findings

Verification errors reduced by up to 22%

8% of users cannot be enrolled and are flagged

DCT-based feature extraction improves discriminability

Abstract

In this paper we simulate a real biometric verification system based on on-line signatures. For this purpose we have split the MCYT signature database in three subsets: one for classifier training, another for system adjustment and a third one for system testing simulating enrollment and verification. This context corresponds to a real operation, where a new user tries to enroll an existing system and must be automatically guided by the system in order to detect the failure to enroll situations. The main contribution of this work is the management of failure to enroll situations by means of a new proposal, called intelligent enrollment, which consists of consistency checking in order to automatically reject low quality samples. This strategy lets to enhance the verification errors up to 22% when leaving out 8% of the users. In this situation 8% of the people cannot be enrolled in the system and must be verified by other biometrics or by human abilities. These people are identified with intelligent enrollment and the situation can be thus managed. In addition we also propose a DCT-based feature extractor with threshold coding and discriminability criteria.