Publication: Development of audio-visual biometric verification system with the integration of fuzzy inference system (fis)
Loading...
Date
2012-06-01
Authors
Tay, Chui Hui
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Biometric verification system is the system that uses physiological and behavioral information to verify an individual. Single biometric verification system shows high reliability in clean noise free condition. However, in the noisy environment, the reliability of the single biometric verification drops drastically. Multi-biometric verification system is developed to deal with the problem faced by single biometric verification system. In this project, voice, visual and audio-visual biometric verification systems are developed. These systems have been evaluated at different noise levels in order to investigate their robustness. In this study, Linear Predictive Coding (LPC) analysis has been used to extract the audio features while Region of Interest (ROI) of lip image has been used as visual features. Support Vector Machine (SVM) is then employed as classifier. The audio-visual biometric verification developed in this project is integrated with the Fuzzy Inference System (FIS). FIS infers the optimum weight to be used as fusion weight in audio-visual biometric verification system. The performances of the audio, visual and audio-visual verification systems are evaluated under various noise levels are verified using the Genuine Acceptance Rate (GAR) versus False Acceptance Rate (FAR) curves. From the experimental results, audio-visual biometric system always outperforms the performances of the single systems. For example at 0.2% FAR, 66%, 69% and 90% of GAR are observed for audio, visual and audio-visual biometric systems, respectively (audio signal at 10dB and visual signal at 0.05 sigma) .