OPTOELECTRONIC RETINAS THAT PERFORM STOCHASTIC GLOBAL OPTIMIZATION

In low-level image processing many problems, such as edge detection, s egmentation, and halftoning, can be formulated in terms of an energy m inimization. When dedicated hardware, namely artificial retinas, are b uilt to perform such tasks at a video rate, only deterministic subopti mal minimization on a limited class of functions is usually carried ou t. We suggest that a suitable combination of speckle and microelectron ics permits the design of stochastic decision-making elements that are able to perform simulated annealing with minimum electronics. Once ma ssively integrated on very-large-scale integrated circuits, these elem ents provide video-rate simulated annealing for an extended class of e nergy functions. This proposition suggests a novel computational strat egy for implementing low-level image-processing tasks.