Abstract

Cellular machines [1,2,3,4] are two dimensional arrays of rather primitive processing elements (PE-s) that are controlled by the internal state of their neighborhood PE-s and (usually) by some external instruction stream in a SIMD style. Such machines offer high computational speed for a useful class of problems, particularly in image processing. While some cellular machines were implemented in silicon, newer versions with enhanced connectivity, such as the cellular hypercube [5], are probably too complex for electronics. Furthermore, the ‘almost-shift-invariant’ geometry of cellular machine interconnections is a good match to optical convolution, which is easy to carried out.

© 1991 Optical Society of America