Measuring vision psychophysically with natural-scene stimuli
The human visual system has presumably evolved so that its component neurons encode the information in natural scenes in a useful and efficient way. Thus, it is argued that the Gabor-like receptive fields of primary visual cortex represent an optimal code for natural visual information. In order to address such a question and to examine how the visual system picks out features from the natural visual environment, it seems to me that we need to attempt psychophysical experiments with naturalistic stimuli (I use monochrome photographs of natural scenes). Ordinary photographs cover an immense variety of different topics and, at first, it seems impossible to believe that experiments on a few photographs could provide general lessons about vision. I will begin the talk by showing that natural images do have much in common and that it is useful to do experiments on a few as representatives. I shall then describe some experiments suggestive that the human visual system is, as proposed, better at making visual discriminations for natural scenes than for unnatural ones. I will describe a simplistic model of information processing in primary visual cortex and how it is capable of explaining many of the experimental threshold values. Lastly, I will describe some experiments in which we used monochrome photographs as masking stimuli - to elevate the detection thresholds for sinusoidal grating test patches. The results are quite paradoxical, but seem to be explained by a model of contrast normalisation in primary visual cortex.