Ergodic HMM-UBM System for On-Line Signature Verification

Abstract. We propose a novel approach for on-line signature verification based on building HMM user models by adapting an ergodic Universal Background Model (UBM). State initialization of this UBM is driven by a dynamic signature feature. This approach inherits the properties of the GMM-UBM mechanism, such as minimizing overfitting due to scarcity of user training data and allowing a world-model type of likelihood normalization. This system is experimentally compared to a baseline state-of-the-art HMM-based online signature verification system using two different databases: the well known MCYT-100 corpus and a subset of the signature part of the BIOSECURE-DS2 corpus. The HMM-UBM approach obtains promising results, outperforming the baseline HMM-based system on all the experiments.

• Type: Conference
• Year: 2009
• Where: COST
• Authors: Enrique Argones-Rúa, David Pérez-Piñar López, José Luis Alba-Castro
• Keywords: COST 2009, HMM-based Online Signature, Multimedia, On-line Signature Verification, Signature Verification