Analogue Architectures for Vision: Cellular Neural Networks and Neuromorphic Circuits

Vision machines based on actual computational methods require the development of simple low-level feature detectors. The low-level feature detectors measure local image properties as scale, orientation, and velocity. Analog VLSI devices that mimic some functionality of biological systems appear to be robust, low power consuming, and fast enough to solve vision problems in real time. In this thesis, it is shown that active resistive di usion networks with low connectivity o er a common framework for the implementation of the low-level feature detectors commonly used in vision (band-pass, wedge, endstopped, velocity-tuned, etc.) yielding to a simple and homogeneous architecture. Di usion networks with four neighbor interactions implement velocity-tuned spatiotemporal lters and oriented spatial lters. Velocity-tuned lters yield to e cient and reliable motion estimation using an analog architecture based on active resistive networks from the photoreceptor level to velocity estimation. Oriented spatial lters using resistive di usion networks yield to a lter basis able to generate complex lters commonly used in vision. From this basis of lters, we generate more complex lters (e. g. oriented quadrature band-pass, quadrature wedge lters) that are approximated by a linear combination of that basis. Changing the linear combination of the basis lters allows the tuning of the architecture to di erent features. The proposed architecture o ers a way to implement both spatial and spatiotemporal lters (motion sensors) with a low cost. This approach opens an issue to the problem of implementing large sets of spatial and spatiotemporal lters tuned to di erent features (edges, junctions, velocity, etc.) in a single chip. Chapter