Mixed-State Auto-Models and Motion Texture Modeling

In image motion analysis as well as for several application fields like daily pluviometry data modeling, observations contain two components of different nature. A first part is made with discrete values accounting for some symbolic information and a second part records a continuous (real-valued) measurement. We call such type of observations "mixed-state observations". In this work we introduce a generalization of Besag's auto-models to deal with mixed-state observations at each site of a lattice. A careful construction as well as important properties of the model will be given. A special class of positive Gaussian mixed-state auto-models is proposed for the analysis of motion textures from video sequences. This model is first explored via simulations. We then apply it to real images of dynamic natural scenes.

• Type: Journal
• Year: 2006
• Where: JMIV
• Authors: Patrick Bouthemy, C. Hardouin, Gwenaëlle Piriou, Jian-Feng Yao
• Keywords: Image Motion Analysis, JMIV 2006, Mixed-state Observations, Observations