Resampling-down Mesh based Discriminant Filter Synthesis for Face Recognition

Abstract

Presented paper addresses face recognition algorithm by means o f synthetic discriminant filters synthesized in pseudo 3-D mesh domain. The objective of the research is to construct facial descriptor in the form of linear filter, which should produce high and low outputs for intra- and inter-class recognition problem respectively. This filter can be synthesized from 3-D sparse meshes derived from a given set of images of a person. Source gray (color) images are subjected to special preprocessing to generate initial mesh. A set ofmeshes for given object class is the basis to construct discriminant filter. As ever the filter is created it is then used as facial descriptor, i.e. serves as personal ID fo r face identification. So far mesh density should be considered as a parameter of the classifier the study examines how the performance of the system depends on mesh sparseness.