In this paper, we present a novel scene interpretation method
by unified modeling of visual context using a hierarchical graphical model.
Scene interpretation through object recognition is difficult due to several
sources of ambiguity (blur, clutter).We model the visual context of scene,
object, and part to disambiguate them during recognition. A precisely
designed hierarchical graphical model can represent the contexts in a
unified way. We also propose a new inference method, particle-based belief
propagation, optimized to scene interpretation in this hierarchical
graphical model. Such an inference method suits the high-level context
of scene interpretation. In addition, our core inference is so general that
it can be used in any complex inference problems. Experimental results
validate the power of the proposed model of visual context to solve the
ambiguities in scene interpretation.