An optoneural system is developed for invariant pattern recognition. T
he system consists of an optical correlator and a neural network. The
correlator uses Fourier-Mellin spatial filters (FMF) for feature extra
ction. The FMF yields an unique output pattern for an input object. Th
e present method works only with one object present in the input scene
. The optical features extracted from the output pattern are shift, sc
ale, and rotation invariant and are used as input to the neural networ
k. The neural network is a multilayer feedforward net with back-propag
ation learning rule. Because of substantial reduction of the dimension
of feature vectors provided by optical FMF. the small neural network
is simply simulated in a personal computer. Optical experimental resul
ts are shown.