IRS: a hierarchical knowledge based system for aerial image interpretation
S. Cosby, Ray Thomas · 1990
A knowledge based architecture for the interpretation of aerial images is presented. The Image Recognition System (IRS) utilises a multiresolution perceptual clustering methodology as a robust alternative to the more traditional edge or region based approaches. Initially, data driven feature generation and primary perceptual clustering is performed independently for two or more reduced resolution versions of the image. A Rule Based Frame System (RBFS) is then used to instantiate more complex geometrical structures from symbolic multiresolution feature representations. Final interpretation is achieved by using knowledge of contextual relations between objects in the domain.