Brightness in human rod vision depends on neural adaptation to the quantum statistics of light

Abstract

In cone-mediated vision, light adaptation obeys Weber's law (constant visual response to constant contrast). Weber adaptation begins in the cones themselves and is well suited to support color constancy. In rod-mediated vision, the visual input consists of a retinal rain of discrete photons. Individual rods do not receive a sufficient number of light quanta within their integration time to light adapt. Instead, light adaptation occurs in post-receptoral circuitry. The nature of this adaptation is still poorly understood. Since the pioneering work of Barlow (1956, 1958), it has been known that psychophysical rod thresholds, when measured with small, brief test probes, increase in proportion to the square-root of adapting intensity, rather than in direct proportion to intensity (Weber's law). A bedrock idea of classical psychophysics ascribes the square-root law to the masking of dim threshold flashes by the naturally occurring variability in the photon rain from the adapting field (Shapley & Enroth-Cugell, 1975). Importantly, this classical explanation does not require actual physiological adaptation and, classically, none was assumed. Here, we present the results of three brightness matching experiments that together demonstrate that the square-root law arises from a neural adaptation (monocular gain control), driven by statistical photon inputs, that influences brightness as well as threshold. The experiments were performed in a haploscope with one eye's adaptive state manipulated and the other eye used as a reference. Our results show that: 1) brightness gain for supra-threshold flashes varies inversely with the standard deviation of the quantal fluctuations from the adapting field; 2) brightness gain is controlled by both the mean and variance of time-integrated photon inputs and changes only slowly (adaptation takes a few minutes). These properties suggest that brightness at low light levels evolved to encode the reliability (signal-to-noise ratio) of sparse statistical light inputs, rather than either luminance or contrast. Meeting abstract presented at VSS 2016