Using overlapping distributions to deal with face pose mismatch - Robotics Institute Carnegie Mellon University

Using overlapping distributions to deal with face pose mismatch

Conference Paper, Proceedings of British Machine Vision Conference (BMVC '05), September, 2005

Abstract

Free-parts representations of the face, based on parametric distributions such as Gaussian mixture models (GMMs), have recently demonstrated benefit in the task of face verification. This benefit can be largely attributed to the representation's natural ability to deal with local appearance variation within the face. Hitherto, a major limitation that has hindered the wider adoption of this type of facial representation, for the task of face verification, has been its poor ability to take advantage of prior knowledge concerning mismatches in context; such as pose (e.g. gallery face=frontal pose, probe face=non-frontal pose). This paper goes some way to alleviating these limitations by making two novel contributions: (i) Demonstrating, via a novel theoretical framework, that a lower-bound can be empirically calculated for how much discriminating information exists for a pre-defined pose mismatch; assuming there is no conditional dependence between observations stemming from different poses. (ii) Through the off-line estimation of subject-independent pose dependent priors, a number of alternatives to the canonical log-likelihood can be employed that enjoy improved performance in the presence of pose mismatch.

BibTeX

@conference{Lucey-2005-9302,
author = {Simon Lucey},
title = {Using overlapping distributions to deal with face pose mismatch},
booktitle = {Proceedings of British Machine Vision Conference (BMVC '05)},
year = {2005},
month = {September},
keywords = {Face Recognition, Pose Mismatch},
}