XnView Software

Is there any advantage to the "pattern" dither that xnview uses? In all my experiments, it looks the worst of any dithering method. Does it have an advantage in computational complexity, or filesize, or something? Floyd-steinberg is popular and seems to work fairly well, plus there is Burkes and a few others.

I don't know what dithering method they use, but paint sho pro has an option to let you choose how much to dither (percentage based). There are also options for median cut and octree.

With few colors, like 2-4, it's often smaller, and has the property of looking better when the nyquist threshold is imperceptible (ie, when the dots are so small they blend together), like print screens, but not on monitors. But for color, Floyd-Steinberg (aka Dot Diffusion) is smaller and generally more pleasant, although it tends to make me think of fax machines. Noise dither looks better still, and several dithers have also been introduced in research (see this paper, for instance) that look cleaner and are often smaller when compressed.

Photoshop has an interesting way of perceptually improving low-res dithers, with a percentage that controls the ratio of dither to thresholding. You'll often find that fewer dots in relatively flat areas looks significantly better than dot spew everywhere, even if it's less accurate. It's nice to have the tradeoff.