Surface mixture models for the optimization of object boundary representation

Abstract

We explore an original approach to represent boundaries of objects based on mixture of densities in parametrized submanifolds embedded in ℝn. This method combines representation of boundary by 'patchwork' of surfaces and traditional mixture models to represent point distributions within the surfaces. Specifically, this method could be used for lossy compression/storage of point clouds, with significant data compression factor. We present method description and experiments with scanned objects. © 2014 Springer International Publishing.