Video modelling and segmentation using Gaussian mixture models

This paper describes a new approach to the video modelling and segmentation problem using Gaussian mixture model descriptors. These have several advantages over the conventional, histogram-based techniques, including: a rigorous statistical basis; the possibility of encoding spatial, colour, texture and motion features in a unified system; and the ability to trade off accuracy of representation against data volume. After a brief introduction to the class of models, results are presented to show their efficacy.