This contribution presents a new class of MRF, that is inspired by
methods of
statistical physics.
The new energy function assumes full-connectivity in the neighborhood
system and thus solves the modeling problems of MRF in the case of
irregular
sites. Moreover, it is defined independently from the considered
application; this
makes it possible to avoid the use of a search algorithm for the
energy minima,
since those and their analytical properties are provided by theory.
Experiments were performed on a database of 12 Objects using
multidimensional
receptive field histograms. We achieved a recognition rate of 98.33%.
