Statistical adaptation and optimal estimation in movement computation by the blowfly visual system

We combine ideas from two lines of research in visual information processing; the study of the fundamental limits to the reliability of selected visual tasks as set by noise at the photoreceptor level, and the attempt to describe certain visual tasks, thought to be of critical importance, in a functional mathematical framework. Movement detection is considered. Instead of assuming noiseless input, we will explicitly incorporate noise in the photoreceptors that provide the input to the computation. For simple movement stimuli such as sudden displacement steps of wide field patterns, this problem is tractable. Comparing measurements of the statistical efficiency of the blowfly movement-sensitive neuron known as H1 with the performance of an ideal observer who uses realistic photoreceptor signals, we find that over a 10 ms time interval the neuron approaches the ideal observer closely. It thus seems that under these simple conditions, the fly tries to compute movement in an optimal way on a behaviorally relevant time scale. One of the questions prompted by this result is whether optimal processing can be generalized to more complex visual tasks, and if so, what role adaptational processes may play in this