Minimal complexity velocity-tuned filters with analogue neuromorphic networks: A theoretical approach for efficient design

In this paper we describe the way for an efficient design of velocity-tuned spatiotemporal filters using analogue neural networks. The filter presented here has a simple velocity-matched structure discrete in space but continuous in time yielding to efficient VLSI realisations and it overcomes some drawbacks of previous similar approaches found in the literature. We show how this filter can be used to compute a distributed representation of velocity similar to that obtained with classical spatiotemporal frequency-tuned Gabor filters.