Probabilistic scene models for image interpretation

Abstract

Image interpretation describes the process of deriving a semantic scene description from an image, based on object observations and extensive prior knowledge about possible scene descriptions and their structure. In this paper, a method for modeling this prior knowledge using probabilistic scene models is presented. In conjunction with Bayesian Inference, the model enables an image interpretation system to classify the scene, to infer possibly undetected objects as well as to classify single objects taking into account the full context of the scene. © Springer-Verlag Berlin Heidelberg 2010.