“Mixed-state Markov Random Fields and Motion Textures”

CRIVELLI T.; CERNUSCHI-FRÍAS B.; BOUTHEMY P.

Mendoza, Argentina

Congreso; AST'2006, “Argentine Symposium on Computing Technology 2006”, Sociedad Argentina de Informática e Investigación Operativa, SADIO, 4 al 7 de Septiembre, 2006, Mendoza, Argentina; 2006

SADIO, Sociedad Argentina de Informática e Investigación Operativa

The aim of this work is to model the apparent motion in  image sequences depicting natural dynamic scenes.  We adopt the mixed-state Markov Random Fields (MRF)   models recently introduced to represent the so-called motion textures.   The approach consists in describing the  spatial distribution of some motion measurements which exhibit   mixed-state nature: a discrete component related to absence of motion   and a continuous part for measurements different from zero.   We propose several significative extensions to this model.   Results on estimation and synthesis from real natural scenes are presented   showing the accuracy and efficiency of the model.  Nota: Versión extendida del trabajo presentado en ICIP'06.