What is Color Banding? And what is it not?

Very well written and illustrated article.

This is one of those things that's so difficult to explain to a mediocre to average graphic designer, much less a layperson. Now the trick would be to get them to read the whole thing...

Dumping color table entries from the last frame of the "256" color GIF:

> gifbuild -d colorBanding_255colorsOn1920pixels.gif | tail -n 200 | grep rgb | sort | uniq | wc -l

> 100

What happened here is some sequence of color space conversions with gamma correction added/removed that ended up compressing the darks.

A true 256 color gray gradient will not display so much visible banding. Just throwing together a naive 256 color gradient in Gimp reveals some RGB triplets with mismatched values suggesting the encoder is doing some odd color space conversion. Even then, it has much less perceptible banding when scaled with nearest neighbor to 1920.

I suspect most of the examples stem from improper gamma handling than any true demonstration of the limitations of 8-bit channels.

Every image on the page is broken for me (Chrome if it matters). I was hoping they would help understand what I was reading. Console is full of errors for where each image tried to load.

The greyscale gradient animations that are meant to show banding are full of gif artifacts so they are not actually smooth gradients.

I was really surprised the first time I saw banding in a 24-bit gradient image, I spent many years thinking it was impossible. But I'm still not sure if that's an inherent limitation of 24 bits, or whether my display is not capable of displaying all 24 bits with full fidelity. The LCD manufacturers are using tricks to squeeze the last couple of bits out of panels that aren't up to the task.

Dithering is seen so often in 90's 256-color images, but why so rarely in the ugly banding of 24-bit RGB images?

That video itself has a lot of color banding, posterized or quantized look. Chrome on 2015 Macbook Pro Retina 13". Same happened on Safari.

I thought this was intentional and expected the author to mention about it at the end of the video, but... nothing.

simply hack pixman and then most things that request gradients in linux (including cairo therefore inkscape) will generate flawlessly smooth color curves

Bit depth is not a measure of quantity, but of dynamic range. In other words, what matters is not only how many quantization steps you have, but how large of a space you are trying to map them over.

The examples in the article point to this obvious conclusion, but don't quite state it explicitly. The relationship between space and quantization effects is demonstrated for a physical dimension interpretation of space by the horizontal greyscale bar that stretches and contracts.

But what if you stretch and contract the dynamic range of your monitor itself? Each bit in the encoding space (naively) offers a doubling of dynamic range in the natural space, so even your 30 bit encoding can be stretched if you display it on a monitor that intends to output a contrast ratio many times greater than what we are used to.

For instance, imagine a monitor that could output light perceptually as bright as the afternoon sun, next to effectively infinite blackness. Will 30 bits be enough when 'stretched' across these new posts of dynamic range, or will banding (quantization) still be visually evident when examining a narrow slice of the space?