algorithmic modeling for Rhino
More dicking around with the median filter. This time each of the RGB channels of an image are populated with 50-50 binary noise, and then iteratively processed by a median filter whose sample window grows from 3x3 to 25x25. The sampling space is tiled in this case, ie. the image has the topology of a doughnut.
The parts where the entire map seems to move right and down coincide with even-sized median windows, as the median of an even number of samples results in a lopsided window.