CONTINUOUS-TIME GLOBAL COMPUTER VISION WITH ANALOG, SPECIALIZED, AND INTERACTING NEURAL NETWORKS

Traditional computer vision considers early vision as an ''inverse opt ics'' problem and tries to invert projective and radiometric equations . It postulates independent modules and uses constraint satisfaction t echniques within each module to obtain the desired inverse. We outline the shortcomings of these approaches and discuss how neural networks can overcome them. We review relevant findings from neurophysiology an d psychophysics and indicate how they have been incorporated into neur al network models. In particular, we stress massive parallelism, nonal gorithmic analog behavior, attention, goal-directed behavior, habituat ion, sensitization, self-organization, and local and global processing properties of neural networks as key elements to analyze visual input s in nonstationary environments.