Abstract

A recognition paradigm is described here which has several characteristics thought to be important in the development of a general machine vision system. A general machine vision system is one which has capabilities similar to human vision. One of the principal features of the paradigm described here is that it learns to recognize objects by seeing them instead of requiring explicit models or rules which have been programmed into the computer. A vision system based upon this principle would be able to interpret visual scenes which fall within its realm of previous experience. Such is also the case in human vision. One can argue, in fact, that development of a general machine vision system will require that the visual memory be learned from its environment rather than programmed by a human. This point has recently been discussed by Haralick and Ulmann.¹

© 1985 Optical Society of America