Abstract
Understanding responses of retinal ganglion cells (RGCs) to natural images is of interest in vision research. RGCs' properties may be different when mapped with natural versus artificial stimuli. For instance, with artificial annuli of homogeneous intensity, surround inhibition of receptive fields (RFs) can reduce RGCs' responses by more than 50% (1). Will surround produce as much inhibition when stimulated by natural images? We attempted to answer this question by recording responses of rabbit RGCs to a large sample of natural images. We used three methods to estimate the level of surround inhibition: First, we estimated linear approximations to RFs with reversed correlation and different calculation methods, including pseudo-inverse, two kinds of regularized pseudo-inverse, and project pursuit regression (PPR) (2, 3, 4). Second, we estimated the Volterra-kernel expansion of RFs, using PPR for dimensionality reduction(5). Third, we segmented natural images into concentric center and surround regions, and studied responses as a function of the mean contrast of the surround. In this method, surround inhibition appeared in a statistical analysis of trend that showed responses falling as a function of surround contrast (6, 7). All these methods showed that surround inhibition was weak for natural images. For example, median responses fell typically by only 10 to 20% over the range of mean surround contrast. Inhibition may be weak, since natural images have low contrasts and have relatively high spatial-frequency components. Functionally, weak inhibition may make sense; although strong surround inhibition helps detecting edges, it may eliminate intensity information from inside objects.
Supported by NIH - Grants EY08921 and EY11170