Accelerated analysis of occlusion

摘要

Instead of relying on heuristics for the automatic analysis of occlusion, this paper introduces provably correct algorithms for simple classes of synthetic pictures. This work is an initial part of research towards efficient picture analysing algorithms that have the same rigour as, for example, parsing algorithms for formal languages. Systematic analysis of a synthetic picture composed of known object images involves a search tree, pruning of which is shown to substantially accelerate the search for feasible interpretations. Quantitative experimental results are reported for noise-free and noisy synthetic pictures.