Three modes of spatiotemporal preprocessing by eyes.

1. Optimal spatiotemporal filters for early vision were computed as a function of signal-to-noise ratio (SNR) and alpha, a parameter defined as the ratio of the width of the probability distribution of velocities as perceived by the naturally behaving animal, and the characteristic velocity of the photoreceptors (the velocity required to move across a receptor's receptive field in a receptor's integration time). Animals that move slowly, on average, compared with the characteristic velocity of their photoreceptors have alpha << 1, animals that move fast have alpha >> 1. 2. For alpha << 1, the temporal part of the optimal filter adapts more to different SNRs (light levels) than the spatial part, leading to large adjustments in temporal resolving power and strong self-inhibition at high SNR, but little lateral inhibition. 3. For alpha >> 1, the spatial part of the filter adapts more strongly than the temporal part, leading to strong lateral inhibition at high SNR, and little self-inhibition. 4. For alpha approximately 1, both spatial and temporal properties change about equally much when varying SNR. 5. Varying the width of the angular sensitivity of the photoreceptors shows that for every combination of alpha and SNR there is an optimal width. Visual systems with large alpha need wider angular sensitivities, in particular at low SNR, in order to reach the information maximum than visual systems with small alpha.