Fancy patterns aren’t totally necessary. A field of random numbers can also provide a pattern with the right properties needed for local partitionless thresholding. Even simple white noise can work, as the header image shows.
But that’s not so pretty. What about other colors of noise? Or minimum wavelengths? The following image shows different colors of noise (white, blue, violet) and different minimum wavelengths (2, 4, 8 pixels).
The effect of the minimum wavelength is very noticeable, resulting in the loss of small details. The noise color is a little trickier. We’ll use grayscale ramps in the following image.
Here we can see that the white noise is spikier, and the violet noise is smoother. This is because the shorter wavelength becomes dominant.
These noise patterns were generated with inverse Fourier transforms, similarly to this post at Red Blob Games.