Arbib Michael A., Hanson Allen R. (eds.) Vision, Brain, and Cooperative Computation

The MIT Press. 1987. — p 705 ISBN-10: 0262510499.For two decades, much of artificial intelligence was dominated by the serial computer, and many problem-solving techniques were structured as if they were inherently serial. However, the field of character recognition, and the field of machine vision, which arose from it, have always been confronted with the fact that the primary input is structured as a vast array representing the stimulus. The analysis of the data in this array is a task of enormous computational complexity. Consequently, many algorithms in machine vision have been logically structured, if not always implemented, in terms of parallel computation. Perhaps for this reason, it is in the field of vision that the interchange between workers in artificial intelligence and those studying the brain has been most fruitful.This volume aims to look at the current state of vision research, stressing contributions from neurophysiology, psychophysics, and computer science. Through it all runs the theme of how best to structure the computations for visual systems. We have adopted the term cooperative computation to designate two different styles of computation. In the low-level vision, the cooperation is in the form of vast arrays of intercommunicating identical processes carrying out such tasks as depth mapping, computing the optic flow, recovering local surface structure, and segmenting the image. At the higher level, processes can no longer be arranged retinotopically (in spatial arrays in correspondence with the two dimensions f the retina or input image) but instead exist in some logical space built on symbolic and relational structures relating to the semantic content of the stimulus. Here the cooperation between knowledge sources, or schemas, is exploited to bring a diversity of knowledge to bear on the interpretation of the various portions of an image, both in terms of the context they provide for each other and in terms of the evolving context of the activity of the organism or machine.