A CMOS silicon VLSI optical sensor

The authors describe a neural network integrated circuit implementation for a pattern recognition sensor that is based on some of the processes of mammalian vision. As in the retina, the image to be recognized is directly projected onto the silicon surface. An array of photoreceptors converts light intensity into currents which are fed into and processed by analog circuit elements. The first layer of image processing detects across the image, the local orientation of lines and edges. This first stage is followed by further layers of processing that recognize more complex patterns in groups of those oriented line directions