Image analysis and segmentation using mixture models

This paper combines statistical modelling with a spatial representation for image analysis. The representation uses the familiar concept of multiple resolutions, but applied to a Gaussian mixture representation of the image: Multiresolution Gaussian Mixture Models (MGMM). It is shown that MGMM can approximate any probability density and can be efficiently computed. After a brief presentation of the theory, examples are used to show how MGMM can be applied to segmentation and motion analysis.