Abstract

Non linear operators have been proved to play a major part in optical computing particularly for neural networks or more generally for Artificial Intelligence procedures involving complex decision rules [1]. Usually a nonlinear operator (NL) gives a zero response for a low input level, a fixed saturated one for a high sufficient value and shows a monotonic transition regime. The latter is generally not considered in the literature for applications in optics, although some recent studies have demonstrated the importance of the degree of the NL characteristic for pattern recognition invariancy, based on template matching classifiers using multilinear pattern decomposition and counting networks [2], Although NL operators are generally the domain of electronics because of higher flexibility, speed and easier handling, electronics is no longer of any interest as soon as such operations have to be performed spatially, which is the case in the latter example. Optical devices exhibiting NL behaviour are numerous [3]. In this paper, we present a solution which seems promising because of its easy handling, its high spatial resolution, the presence of different NL phases as well as the possibility of adjusting these NL characteristics, hence allowing, in the short term, the evaluation of many optical implementations of complex procedures such as described in [2].

© 1990 Optical Society of America