Image interpretation by distributed cooperative processes

Special-piirpose vision systems have shown considerab le success within limited task domains, b u t to d a t e t h e r e are n o general-purpose systems working effectively across a variety of task domains. T h e VISIONS Schema System’ provides a framework for building a general in te rpre ta t ion sys tem as a distributed network of m a n y small special-purpose in te rpre ta t ion systems. Each schema is a n “expert” a t recognizing one t y p e of object. Schema instances r u n concurrently, communicating asynchronously through a global blackboard, each trying t o establish t h e presence of i ts particular object b y invoking general purpose knowledge sources. I n th i s way schema instances cooperate a n d compete t o identify a n d locate t h e significant objects present i n t h e scene. This paper discusses t h e problems of knowledge representat ion i n a distributed AI environment , a n d t h e Schema System’s approach t o those problems. A series of interp re ta t ion exper iments have been performed o n nine images f rom t w o n a t u r a l scene domains; results f rom t h r e e images will be presented here. In te res ted readers are referred t o [6] for a full description of t h e sys tem and t h e experiments .