Handwritten zip code recognition using an optical radial basis function classifier

摘要

A new optical architecture is described for the realization of distance-based classifiers and, in particular, radial basis function classifiers. This new system uses spatial multiplexing to improve computation speed while simple parallel postprocessing electronics are used to provide on-line learning in the optical neural network. This architecture is compared with a previous optical system which realized a radial basis function classifier in the solution of the handwritten digit recognition problem.