A fundamental characteristic of computer vision is its unobvious modularity. In that sense it is truly an artificial intelligence discipline. In a recent article (Scientific American, September 1995, special issue on "Key technologies for the 21st century"), Doug Lenat writes: "It is the prerequisite knowledge, not the content of the text" (the article is about natural language understanding),
"that [has] to be codified". It is worthwhile to note that in computer vision, while it is important to represent the prerequisite knowledge, an equally major additional issue is what information is available in the sensory data (images), which turns out to be as difficult a task as codifying the prerequisite knowledge. The representation and use of knowledge and the image signal are intertwined in a complex manner which have been found hard to decouple.
Recently, world chess champion Garry Kasparov was defeated by a computer program. This is clear evidence that a major progress has been made in representing and exploiting domain specific knowledge in artificial intelligence. However, a system with broader capabilities remains to be a target still under pursuit. Analogously, there have been a number of vision programs which could be considered true successes. These have addressed basic, theoretical foundations as well as problems in specific, applications domains. The former are exemplified by the work on a variety of approaches to threedimensional interpretation and recognition. The latter are the likes of chess programs; e.g., a vision program could be written to recognize the pieces of a given chess set. However, the achievements of both types notwithstanding,
it remains an open problem to evolve and test from the basic efforts a theory of computer vision, and similarly, to extend the solutions in one specific domain to other domains. One reasonable way is to continue efforts in both directions but consciously emphasize extension and generalization.
The work on increasing the scope of the basic modules of three-dimensional interpretation