Abstract
Spatial contrast sensitivity varies with ambient light levels and eccentricity (Shapley and Enroth-Cugell, 1984) and is thought to be largely determined by the retinal code. Light adaptation of contrast sensitivity was previously shown to match theoretical predictions of efficient coding in the retina (Atick & Redlich, 1990; van Hateren, 1992). However, the actual adaptive change is more complicated and depends on both light adaptation and visual eccentricity (Koenderink et al., 1978). Here, we show these phenomena can be predicted from the same theoretical principle. Specifically, we employ a generalized model of optimal neural coding that, unlike previous models, can be optimized for any number of neurons relative to the input dimension and counteracts inherent noise and signal degradation (Doi and Lewicki, 2011). We model 1) optical blurring and photoreceptor noise, 2) the local convergence from cone photoreceptors to retinal ganglion cells (RGCs) which varies dramatically with eccentricity (Goodchild et al., 1996; Ahmad et al., 2003), and 3) the limited neural capacity of RGCs (Borst and Theunissen, 1999). We find that the adaptive change of model RGC receptive fields is larger at the fovea as in the previous studies, but becomes progressively smaller in the periphery. This is consistent with available neural data (Barlow et al., 1957; Enroth-Cugell and Robson, 1966), although experimental data in the fovea, where the predicted change is greatest, is lacking. Finally, we show that the optimal retinal code can be used to predict psychophysical light adaptation and contrast sensitivity and is consistent with experimental data at the fovea and different eccentricities (Koenderink et al., 1978).
Meeting abstract presented at VSS 2014