الفهرس 
Introduction
Future directionOnly 14 pages are availabe for public view
 |  | from | 169 |  |  |
| | | from | 169 | | |
Abstract
Biometrics can be defined as measurable characteristics of the individuals based on their physiological features or behavioral patterns that can be used to recognize or verify their identity. Biometric technologies attempt to automate the measurement and comparison of such characteristics for recognizing individuals. Many different technologies have recently been developed for person recognition and identity authentication. Some examples include measures based on information from handwriting, fingerprint, face, voice, retina, iris, hand or ear shape and gait data.
A practical biometric system should meet the specified recognition accuracy, speed, and resource requirements, be harmless to the users, be accepted by the intended population, and be sufficiently robust to various fraudulent methods and attacks to the system. Previous requirements are satisfied using a multimodal audio-visual based system that includes the speech signal and the face image signal. Acoustic-based methods are susceptible to low acoustic signal-to-noise ratios or channel distortion. Similarly, face based recognition performs poorly when presented with pose/ illumination/expression variation and occlusion. The audio and face modalities contain non-redundant complementary information regarding person identity. A signal degradation factor is unlikely to affect both audio and video signals. For these reasons, to combat unimodal limitations, a multimodal signal fusion approach is adopted to improve both robustness and overall person recognition performance.
In the present work, a multimodal biometric system which integrates face image and speech data to make a personal identification is considered. Our choice of these two specific biometrics is based on the fact that they have been used routinely in law enforcement community. Further, these biometric indicators complement one another in their advantages and strengths.
The proposed system is targeted for verification applications to authenticate the identity claimed by a user such as in a multi-user account authentication. It consists of four components; acquisition module, template database, enrollment module, feature extraction and verification module using a feature fusion method and an artificial neural network. Several contributions to the field of audio-video person recognition and multimodal signal processing resulted from the work described in this thesis.