A Real-Time Web-Based Personal Authentication System with Visemes and Acoustic Biometric Features

• Main Authors: Fu-Rong Cheng, 鄭復榕
• Other Authors: Kuo-Chin Fan
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/31792083547696283262

碩士 === 國立中央大學 === 資訊工程學系碩士在職專班 === 93 === Phonemes are the basic acoustic elements of a language, and the visemes are the lip shapes of a word while speaking. Since the uttering characteristic/manner is unique to each individual [8] [9], a personal authentication system can fully make use of the biometric information. The biometric uttering features include lip visual features [1-3], visemes [4], and phonemes [4] of an individual. In this thesis, we propose an effectively web-based personal authentication system by utilizing filter model with biometric uttering features in Chinese, that is a user equipped with a web camera can register and log in the system later by the registered biometric uttering features. In our work, we compare and analyze these biometric features, then build 2 GMM classifier, and finally fuse the outputs to a personal authentication system. The consideration of system scalability and networking bandwidth may more or less sacrifice the recognition rate. However, experiments results shows that the proposed system can handle data stream of personal frontal view in 320*240 pixels with 30*15 pixels of lip in reasonable response time and satisfactory recognition rate above 80% can be achieved.