On the basis of a fundamental description of a biological visual system and some psychological experimental results on image transformation in brain, a field theory of pattern identification is given. In this system the image in brain is postulated to be freely transformed into the standard one for pattern identification by the excitation of gauge field. Taking account of the idea of the image transformation, an optical-electronic recognition system is proposed as a model of biological visual perception. In the optical feature extracting part, the planar microlens array is used to form many images of an input pattern simultaneously, and from each image feature extraction is performed in parallel by the optical correlation with orthogonal spatial filters. The planar-lens array followed by spatial filters works as the hypercolumn structure in the first visual area in brain. In the recognition part, the extracted features are reorganized in the neural network such as "perception" or "neocognitron", giving rise to an image of input pattern. The reorganizing process corresponds to that in higher visual area in brain. For realizing the optical feature extracting part of the proposed system, some basic experiments on miniaturized matched spatial filter (MSF) are reported. In this system a graded index lens of small diameter 3.3 mm is used. The partial geometrical features of input pattern of the small size lx1 mm are shown to be extracted through the optical correlation with the miniaturized MSF.
"Multimatched Filtering System As A Model Of Biological Visual System", Proc. SPIE 1014, Micro-Optics, (5 April 1989); doi: 10.1117/12.949437; https://doi.org/10.1117/12.949437