These days, portable personal devices such as PDAs or mobile phones are indeed widely used. They provide the mobile worker or the customer with portable computing and wireless access to Telecom networks and to the Internet. It is then possible to provide anywhere anytime a natural access to any service, such as PIN code replacement, phone card reloading, remote purchase, telephone banking or voice-mail. Most of these services involve micro payments that can currently be done only using PIN codes or passwords.
To win wide consumer acceptance, their friendly, personalised, interactive interfacesmust recognise people in their immediate environment and, at a minimum, know who they are. The conventional means of identification such as passwords, secret codes and personal identification numbers (PINs) can easily be compromised, shared, observed, stolen or forgotten. In view of this, it appears that the required optimal reliability in determining the identities of users may only be achieved through the use of biometrics (automatically recognising a person using distinguishing traits). Since more and more portable devices are equipped with a microphone and a video camera (while very few devices are equipped with fingerprint or iris scanners), MoBio will thus focus on multiple aspects of biometric authentication (ranging from research to development and scalability) based on face and voice authentication.
Starting from the state-of-the-art systems available from the MoBio partners, the goal of this project is thus to further study, develop and evaluate bi-modal (face and voice) biometric authentication (BMBA) technologies in the context of portable and networked devices. Although biometric authentication is a complex problem, and is still not reliable enough to be widely accepted, it has also been shown that the use of multiple modalities increases the performance of biometric systems. However, most of the current multi-modal biometric systems simply perform fusion (of the outputs resulting of the independent processing of the modes) and do not actually take advantage of temporal correlations between modalities. As a matter of fact, very little work in the research community has been done on joint multi-modal fusion to perform joint authentication of several modalities (in our case face and voice).
In MoBio, we shall carry out research on joint bi-modal biometry under various realistic conditions. More precisely, this project will investigate the following technologies: robust face localisation and speech segmentation in noisy environments, video-based face authentication (in order to avoid replay attacks using pictures of the face we should perform face authentication over the video), speaker authentication, bi-modal authentication (both expert fusion and joint face/speaker authentication to take full advantage of the correlation betweenmodalities) and unsupervised model adaptation thought time. MoBio will thus address several innovative aspects in the framework of mobile devices, including:
1. Advanced research and development on joint bi-modal authentication (as opposed to bi-modal fusion), involving the development of new statistical models actually processing both channels simultaneously and in a principled way.
2. Investigation of model adaptation techniques to reduce the degradation of biometric systems over time,
3. Analysing the scalability of the proposed solutions by studying how the performance of the system degrades while the complexity of the model is reduced.
4. Providing common evaluation tools and baseline results to the research community in order to evaluate and compare the developed technologies.
The project will also address the development of a demonstration system. We will investigate two main scenarios:
• Embedded biometry where the BMBA system is running entirely on a mobile phone. The system is designed to maximise the authentication performance and to minimise resources such as CPU, memory and speed.
• Remote biometry if the BMBA system needs too many resources to reach the required performance it will be hosted on a server while a minimum of essential functionalities would stay on the mobile phone such as capture, segmentation, preprocessing and feature extraction.