System architectures for interactive knowledge-based image-interpretation

Abstract
We discuss hardware and software architecture for automated image-interpretation. The importance of considering the complete system is emphasized leading in particular to the conclusion that high-level and low-level processing are intimately linked. We present arguments to support the idea that automated image-interpretation systems should be knowledge-based and interactive. We attempt to identify the main architectural problems which such systems must address and outline a systematic strategy for acquiring, structuring and using knowledge.