An artificial retina for SAR object recognition

A conceptual model of an artificial retina has been developed for object recognition in synthetic aperture radar (SAR) imagery. Candidate strategies for the development of mechanisms within the artificial retina were derived by comparing and contrasting principles that underlie the human visual experience, from the statistical behaviour of SAR returns, and from published studies of mammalian neurophysiology. From these studies, a nonlinear analogue of center-surround difference operators observed in the human retina was developed that exploits unique statistical properties of SAR formed image returns in the log-magnitude domain. This operator forms the basis for retinal mechanisms for amplitude compression, gain compensation, frequency channelization, and focus-of-attention. The union of these mechanisms yields the complete artificial retina. We introduce the architecture, describe its major mechanisms, and present the results of recognition experiments in which the artificial retina is incorporated into an existing object recognition system.