it's a way of trading off (effectively) frequency response to gain signal to noise ratio at a lower bandwidth.
It is widely used everywhere, from audio ADC/DACs (sigma-delta modulators), and images naturally have dithering due to noise which makes them look pretty nice despite the limited bit depth of 8 bits.
Dithering is useful if you can trade samples for accuracy. It is useful if your quantization scale is too coarse.
As a result, one can inject noise to sense signal that is normally below a sensor threshold, for instance.
In graphics, it is still very useful to add some dithering in some cases; one obvious example being to reduce banding on gradients: https://en.wikipedia.org/wiki/Colour_banding
Regardless of the chosen quantization scale, there are corner cases where values measured will fall close one to another; dithering is useful in these cases, and is not limited to images, but any data: https://en.wikipedia.org/wiki/Audio_bit_depth#Dither
Dithering is super useful in all sorts of applications where you need to record an analog signal into a digital format. If you don't dither you can get weird sampling distortions.
If you have a undithered shallow gradient from black, it will look terrible as you can clearly see bands where the pixels go from #000000 to #010101 (or whatever color the gradient goes towards) and further steps. We had one of those gradients in an app where an image faded out to black, it looked great. And then on the next phone model, the black background was really black instead of a dither pattern and it looked absolutely terrible.
It's hard to accurately drive e-paper, so some prefer to consider them black-and-white screens. However, taken at a macroscopic level, it has quite an important bit-depth, as while each individual particle might have only ~two possible states, you are controlling a distribution of more than a hundred of these per pixel.
In a sense, the variability (particle size, etc) will already introduce some noise, which, for instance, makes it so that you need less antialiasing.
Another use of that variability is that those screens can effectively be driven as if they had a very deep dynamic range.
Your statement would be true if all particles were strictly the same, reacted at the same voltage, had exactly two equilibrium points, and didn't interact with each other. Luckily, reality isn't so dull, and you don't really need dither if you put a bit more effort into controlling the voltage being applied: https://news.ycombinator.com/item?id=16140284
Low Tech Magazine's solar powered site uses dithering to help achieve very low power consumption. I think the results look awesome, just as good as the artfully filtered images used by a lot of sites but with a fraction of the weight.
A lot of monitor these days have local/micro dimming in order to provide decent contrast and brightness. It's especially true with HDR displays. So there is a high chance that the dithering pattern trigger one or more dimming zone of your monitor.
18 comments
[ 2.9 ms ] story [ 49.2 ms ] threadHere's a couple more dithering algorithms: https://bisqwit.iki.fi/story/howto/dither/jy/
It is widely used everywhere, from audio ADC/DACs (sigma-delta modulators), and images naturally have dithering due to noise which makes them look pretty nice despite the limited bit depth of 8 bits.
Dithering is useful if you can trade samples for accuracy. It is useful if your quantization scale is too coarse.
As a result, one can inject noise to sense signal that is normally below a sensor threshold, for instance.
In graphics, it is still very useful to add some dithering in some cases; one obvious example being to reduce banding on gradients: https://en.wikipedia.org/wiki/Colour_banding
Regardless of the chosen quantization scale, there are corner cases where values measured will fall close one to another; dithering is useful in these cases, and is not limited to images, but any data: https://en.wikipedia.org/wiki/Audio_bit_depth#Dither
However it can be used in other contexts, like in video.
In a sense, the variability (particle size, etc) will already introduce some noise, which, for instance, makes it so that you need less antialiasing.
Another use of that variability is that those screens can effectively be driven as if they had a very deep dynamic range.
Your statement would be true if all particles were strictly the same, reacted at the same voltage, had exactly two equilibrium points, and didn't interact with each other. Luckily, reality isn't so dull, and you don't really need dither if you put a bit more effort into controlling the voltage being applied: https://news.ycombinator.com/item?id=16140284
https://solar.lowtechmagazine.com/2018/09/how-to-build-a-low...
As someone who grew up in Sweden, Futhark to me is a runic alphabet :) https://en.wikipedia.org/wiki/Elder_Futhark