I was looking at the differences between the image I thought I should project, and the image actually projected. But I forgot that the latter had been rounded to integer values, so I got a pattern of errors ranging from -0.5 to 0.5.

I don’t remember how I got this image, but I like it.

This is the basic sort of image used in a lot of my (and others!) research: a binary random-dot pattern, in which each pixel is colored either black or white at random. Back in the 1960s, Bela Julesz gave a huge impetus to research into binocular vision when he demonstrated that humans can perceive depth in stereograms made from such images.