PDF version of this entire document

Rationale

The merit of the approach adopted here is that it goes beyond 2-D and makes use of the full 3-D data, relying on PCA to handle an otherwise very complicated job³. One of the current pitfalls when it comes to 3-D face recognition is the inability to manage a lot of data and exploit its full potential; in 2-D there is also some guesswork associated with uncertainty, caused in part by illumination ambiguities, which only ever allow rough estimation of depth and never a consistent method either, due to changes in light sources and scale factors. 3-D face images suffer from none of these issues, but they are more intrusive in acquisition, not to mention secondary matters relating to expense, storage limitations, and availability.

This document provides some background about a research project which deals with a fully 3-D face-related application. A clear direction has neither been determined nor finalised yet for building upon this work, but suggested improvement might be the utilisation of GMDS. The group of Dr. A. Mian (http://www.csse.uwa.edu.au/%7Eajmal/home page) has done some fantastic work on 3-D face recognition and we shall attempt to reproduce some results with a NIST-supplied database, then show potential for substitution and possibly an improvement (performance- or detection-wise, where by performance we refer to speed and hardware utilisation).