Mid-Level Vision Processes for Automatic Building Extraction

摘要: Mid-level processes in vision are understood to produce structured descriptions of images without relying on very specific semantic scene knowledge. Automatic building extraction can use geometric models a large extent. Geometric hypotheses may be inferred from the given data 2D or 3D and represent elementary constraints as incidence collinearity more relations symmetries. The hypothesis lead difficulties during spatial inference due noise inconsistent mutually dependent constraints. paper discusses selection not-contradicting via robust estimation set independent prerequisite for an optimal objects shape. Examples analysis image range given.