
Title
Mixed-state spatio-temporal auto models
Speaker
Cécile Hardouin, Université Paris Ouest
Abstract
We consider a general model for mixed-state spatio-temporal data. Our data consist of two different types: The observations record many zeros, along with continuous real values. Such data occur in many applications, like rainfall measurements or motion analysis from image sequences. We present a spatio-temporal model for these data, namely a Markov Chain (in time) of Markov random fields (in space); the Markov random fields are generalized Besag auto models for mixed-state data. We present an application to spatio-temporal data of motion textures from video sequences.