MUST: Multimodal User Authentication for Secured Transaction

• Research unit: COST
• Project number: C01.0120

Key words (English)

Authentication; face verification; seaker verification; fusion

Research programs (English)

COST-Action 275 - Biometrics-based recognition of people over the internet

Short description (English)

See abstract

Further information (English)

Full name of research-institution/enterprise: Institut Dalle Molle d'Intelligence Artificielle Perceptive IDIAP

Partners and International Organizations (English)

BE, HR, FR, EL, IE, IT, PL, PT, SK, SI, ES, SE, CH, UK

Abstract (English)

MUST: Multimodal User Authentication for Secured Transaction The main objective of MUST is to investigate effective methods for the recognition of people over the internet based on voice and facial characteristics in order to facilitate, protect and promote various financial and other services over this growing telecommunication medium. During the last year of the project, IDIAP has concentrated on several aspects of multimodal authentication for secured transactions, mainly related to speaker verification, face verification, and fusion of modalities. Speaker verification: Based on the previously proposed discriminant framework, and following recent advances in kernel methods, we have developed novel large margin kernel based text independent speaker verification systems that obtains state-of-the-art results on benchmark databases. Face localization and verification: We have compared state-of-the-art discriminant approaches to face verification with generative based approaches inspired from speaker verification systems, and the latter, with appropriate adaptation for the face verification domain, appear to yield more robust and better performance on various benchmark databases. We have also proposed several new features to be lighting and condition independent. Fusion of modalities: We have proposed several benchmark protocols and databases to better compare fusion algorithms in the context of multimodal authentication. We have also proposed several theoretical analysis of the state-of-the-art fusion models.

References in databases (English)

Swiss Database: COST-DB of the State Secretariat for Education and Research Hallwylstrasse 4 CH-3003 Berne, Switzerland Tel. [phone] Swiss Project-Number: C01.0120