Video of a mongrel synthesis:
In this video, we have extracted summary statistics from a black and white checkerboard. We then use a version of Portilla and Simoncelli (2000) to synthesize a texture with approximately the same summary statistics. The algorithm works by starting with a random image -- here, random noise -- and iteratively imposing statistical constraints. The video shows the algorithm iterating until the synthesis converges. We have artificially forced the early iterations to remain onscreen for longer; otherwise, it would be difficult to see the early steps of the algorithm due to the rapid convergence.
These summary statistics form the basis of our model of peripheral vision. The synthesis algorithm enables intuitions about the model, as well as empirical testing of the model.