Image statistics in the visual system

The overall aim of our project is discover and describe statistical calculations performed by the human visual system.Our perceptions are thought to result from the combination of prior knowledge with data we gather from the environment. Of course, the visual system cannot afford to be too meticulous a scientist and measure everything. Real environments are just too complex and dynamic. Instead, the visual system must settle for being a clever statistician and quickly estimate various visual features.Exactly which visual features can be quickly estimated is a matter of some debate, but there is no question that spatial orientation (i.e. tilt) is one of them. The primary evidence for orientation's special status in the visual system comes from search experiments, in which observers must find a target region, having a unique orientation, amongst distractors, having different orientations. When the average orientation of the target region is sufficiently different from that of the distractors, then the number of distractors will be irrelevant; the target just pops out. Thus, we know that the visual system can compute mean orientation without too much conscious effort. What about orientation variance? That is, would a region with lots of different orientations pop out from regions with largely similar orientations? If the visual system proves capable of this type of statistical analysis, can it do something similar with other visual features, like colour and size?