Multiresolution Gaussian mixtures for image analysis

This paper introduces a generalization of scale-space and pyramids, which combines statistical modelling with a spatial representation. The representation uses the familiar concept of multiple resolutions, but applied to a Gaussian mixture density representation of the image. It is shown that MGMM can approximate any probability density and can accommodate the effects of smooth motions. After a brief presentation of the theory, examples are used to show how MGMM can be applied to problems such as segmentation