Integrating Monolithic and Free-parts Representations for Improved Face Verification in the Presence of Pose Mismatch

Abstract

Face images, varying under pose, are dramatically different in their ?ixel?appearance even if they stem from the same subject. Our work concentrates specifically on the task of verifying faces when the gallery set stems from frontal face images, with the probe set stemming from a number of alternate poses (i.e. pose mismatch). An argument is put forward for attempting to recognize faces through integrating holistic/monolithic and free-parts representations. Canonical monolithic representations are investigated such as Eigenface and Fisherface techniques, as well as recent techniques that are able to deal with pose specifically, such as Eigen-light fields. Similarly, parts representations are investigated, with particular attention being paid to Free- Parts Gaussian Mixture Models (FP-GMMs) as a useful representation. A contribution is made via the analysis of what traits, in a face, are most useful for each representation. Finally, we are able to demonstrate that there is: a) benefit in combining free-parts and monolithic representations, and b) further benefit can be obtained by varying the weight placed on each representation as a function of view point.