Analysis by Synthesis: 3D Image Parsing Using Spatial Grammar and Markov Chain Monte Carlo
Siyuan Qi · eScholarship (California Digital Library) · 2015
Scene understanding is a fundamental problem in computer vision research. Weaddress this problem in an “analysis by synthesis” fashion - explain observed data(an 2D image) according to a set of spatial grammar (describes the underlyingfunctional arrangement and 3D geometric structure of a scene) that generate it.The inference process is carried out in a Bayesian framework. The posteriorprobability includes a prior probability reflecting the knowledge of indoor 3D scenestructure encoded by grammar, and a likelihood that evaluates the accuracy of there-projected image and the physical plausibility. The most reasonable explanationof the image is given by a parse tree that maximizes the posterior probability, andit is found by reversible-jump Markov Chain Monte Carlo sampling.