Optical neural network having a function of relative feature extraction without inhibitory connections

An optical neural network (ONN) model having a function of feature extraction without any inhibitory synaptic connections has been proposed. The model called relative feature extracting ONN (R-FEONN) can recognize handwritten alphabet and numerical letters by extracting relative features which are hardly affected by changes in size or shifts in position of the input pattern. The optical system realizing the feature extraction is described.