Partenaires et organisations internationales
(Anglais)
|
MatraNortel (F), Cerberus AG (CH), Ibermatica SA (E), EPFL (CH), University of Neuchatel (CH), UCL (B), University of Surrey (UK), Renaissance (B), Aristotle University (GR), Compagnie Europeenne de Telesecurite (F), Universidad Carlos III (E), Banco Bilbao Vizcaya (E), Unidad Tecnica Auxiliar de la Policia (E)
|
Résumé des résultats (Abstract)
(Anglais)
|
The goal of the M2VTS project is to address the issue of secured access to local and centralised services in the multimedia environment. The main objective is to extend the scope of application of network-based services by adding novel and intelligent functionalities, enabled by automatic verification systems combining multimodal strategies (based on speaker recognition, face recognition, and classifier combination techniques). Biometric authentication techniques like face recognition and speaker recognition are non-intrusive and therefore more acceptable by the user than intrusive methods such as finger-print recognition or retina scans. However, the performance of face- and speech-based recognition techniques is usually lower than for other methods. These methods therefore often don't meet the high performance requirements imposed by typical applications. One objective of the project is to show that multi-modal systems can overcome the shortcomings of mono-modal systems, taking advantage of the emerging multimedia environment.
The major achievements of the project can be grouped into four categories: the recording of large multimodal databases (M2VTS, XM2VTS, LoCoMic), the development of multimodal verification techniques (face recognition, speaker recognition), the development of hardware platforms (DSP), and the realisation of several applications for secured access (cash dispensers, buildings, tele-services).
IDIAP has made the following contributions to the project: We have developed a new method for person authentication that simultaneously combines the acoustic speech signal with visual motion information of the face while the person is talking. Two speaker verification methods were developed, a text-dependent and a text-independent system. We have also investigated different strategies for the combination of classifiers, including support vector machines, multi-linear classifiers, and logical analysis of data. Our monomodal and multimodal systems have shown outstanding performances within the consortium when tested on the XM2VTS database. Furthermore, a face detection system has been developed to allow the unconstrained used of face recognition systems. We have also implemented a PC-based speaker verification demonstrator and have collected a speaker verification database under real-world conditions. One of our verification algorithms has been implemented by Ibermatica SA and by Cerberus AG for application demonstrators.
|
