A new multimodal biometrics for personal identification using machine learning techniques

Publication Type:
Issue Date:
Filename Description Size
Thumbnail01Front.pdfcontents and abstract5.48 MB
Adobe PDF
02Whole.pdfthesis75.49 MB
Adobe PDF
Full metadata record
NO FULL TEXT AVAILABLE. This thesis contains 3rd party copyright material. ----- Multiple biometrics are used to compensate for the limitations impinging upon unimodal biometrics. Many fusion or combination techniques of different biometrics have been proposed for personal identification and verification in the past. Among many biometric characteristics, the facial biometric is considered the most un-intrusive single modal technology that can be deployed in the real-world visual surveillance environment. These uni-modal technologies suffer from many variation problems such as pose, illumination, or facial expression due to the real-world unconstrained environment. Considerable research has been done to cope with the variations due to poses and lighting conditions, and multiple biometrics (such as gait and ears) which can be integrated with facial biometrics have been proposed to compensate recognition performance when recognizing faces at a distance. However, little research attention has been paid to facial expression changes. In most literature, facial expression changes are considered as noise that would degrade the recognition performance. However, can these intra-personal variations be used as another behavioral biometric and also be useful for assisting the extra-personal separation to improve personal identification performance? Our hypothesis is that the dynamic information of intra-personal facial behavior could be used not only as another behavioral biometric but also could assist the extra-personal separation for recognition perfonnance improvement. We will propose and design various experiments to validate and to support our hypothesis. We firstly discussed the facial expression variation problem, secondly, we introduced another single behavioral biometric using facial behavior, and finally, we proposed a framework to integrate facial appearance and facial expression features for improving the personal identification performance. Our experimental results showed that facial behavior can not only be used as another behavioral biometrics in single modalities, but also can assist in extrapersonal separation in multiple modalities for personal identification improvement.
Please use this identifier to cite or link to this item: