ABSTRACT
The long-term goal of the machine vision effort is to synthesize a general purpose system that can perceive and understand images of an unconstrained environment in ’real-time.— Our near-term goal is to understand how to specify a general purpose vision (GPV) system modeled after human vision. We assume that all of the visually guided human behavior can be realized in terms of a finite kernel of visual tasks. We attempt here to (a) identify these tasks, (b) translate them into routines and algorithms, (c) relate these routines to functional modules as constrained by information from neuroscience, (d) specify connectivity between modules, (e) synthesize functions computed by the modules in terms of neural structures.
