Fast plausibility checks of image interpretation maps

Abstract

The semantical interpretation of images is usually done by AI functions. Though they deliver mostly good results, errors do occur rather often. This motivates adding some functionality to check the plausibility of the interpretation results. Such checks may be formulated by a set of probabilistic rules which assign probabilities to geometrical relations found in the interpretation. However, the evaluation of a system of rules may become computationally expensive. We propose to precompute the evaluation and store the results in an optimized form. Namely, we use Algebraic Decision Diagrams to represent the results of the plausibility check. The decision diagrams provide the plausibility check result at query time at a low computational cost.