Figure 9.

A simple model of missing pixel estimation in humans. The input image is passed through a center-surround linear filter and normalized by the local luminance, producing a Weber contrast response at each pixel location (similar to the response of a ganglion cell). The estimate of Weber contrast at the missing pixel is obtained by applying a fixed set of linear weights (shown on the right) to the eight pixel locations immediately surrounding the missing pixel. These weights were learned on a training set consisting of millions of randomly selected natural image patches.