Optoelectronic cellular automata for video real-time vision

In this work, we elaborate on the compromise between the efficiency of a multiprocessor computer architecture for handling large classes of computing tasks and the good performance of optics for implementing shift invariant operations, in particular convolution. We derive a class of processors, optoelectronic cellular automata, that can efficiently implement intensive, low level vision tasks in a time compatible with application constraints up to standard video rates. As one illustration, a parallel simulated annealing task performing motion detection on an image sequence is demonstrated.