Organic Vision

Computer vision is difficult, because you either have it all or you have none of it.

The activity of computer vision to date can to a very large extent be characterized as a collection of efforts to develop individual algorithms dedicated to dealing with individual information sources such as edge finding, motion extraction, stereo fusion, shape from shading and so on. One by one these efforts have largely been frustrated. Edge finding is a case in point: after 40 years of effort it is an unsolved problem.

The realization is dawning on the field that individual algorithms cannot satisfactorily solve individual problems. Each one of these efforts runs into tremendous ambiguities (except in very special, fortuitous visual situations). These ambiguities are traditionally dealt with in PhD thesis projects by hand-tuning parameters to the needs of a given, narrow selection of sample images. Only coordination of a large number of modules, each subserving a single source of information, can reduce the uncertainties in parameter settings for each one of them. Visual scenes can only be analyzed reliably by the coordination of subsystems.

The scientific program resulting from this is Organic Vision: Attention must be shifted from development of individual algorithms to the organizational mechanisms required to coordinate many of them. This sets the stage for a collaborative effort among many laboratories, to wrap existing algorithms in standardized interfaces, make them freely available to the community as a whole, and concentrate efforts on the issues of how to coordinate these subsystems.

A third of our brain is dedicated to vision. This makes it more complex than any man-made single software or hardware system so far! The development of large software systems, such as computer operating systems may today run up to $1 Billion. Computer vision would be much more expensive than that if it was to be developed in current algorithmic style. Only new, vastly more efficient system development techniques will be able to make technology see in natural environments. Systems must be grown and trained, not written. No vision but organic vision!

Soon, the appearance of camera-plus-processor units of $10 each may create a tremendous market pull to infuse computer vision with new energy. As far as information technology goes, computer vision may then turn out to be the pace-maker of Organic Computing.