Error reduction during visual processing by the retina - a paradigm for sensor design

We use a general second-order mathematical description of sensing, sensory information acquisition and pre-processing to treat biological vision in analogy to a technical sensor. Using very early visual processing by the retina, i.e. the retinal receptive fields, we compare the structure of interconnection of photoreceptors, bipolar and ganglion cells that evolved in vertebrates with other structures of interconnections that would serve the same purpose of generating an image of contrasts and supplying it to the brain. We are able to show that the ladder-type structure of the vertebrate retina approaches ideal behaviour (with respect to high linearity and low error sensitivity) and is the structure most tolerant to imperfection of its constituent elements, the different nerve cells. Since the model presented is very general, it can be applied to other sensory systems as well. The application of the sensory structure of the retina (which, from the biological modelling point of view, derives in a straightforward way from known facts about retina physiology) is useful if applied to sensor design. In particular it shows how nonlinearities, error- and cross sensitivities of the entire sensor system can be reduced by the structure among its elements.