4:4:4 RGB correlation?

Max Field · December 8, 2019

I can't remember where, but I heard that each of the 3 numbers in the chroma subsampling is a representative of a different tone in the RGB colorspace?

If it's 422 it goes like 4=G:2=R:2=B ?

Is this how it works or is it more correlated to how the colors will lose resolution evenly from a geometric perspective?
Thanks

Phil Connolly · December 8, 2019

Nope - 4:4:4 maps to R:G:B

But 4:2:2 is Y:U:V

Y = is luma (or black and white image)

U and V are colour difference signals (not actual colours) RGB can be mathematically derived from Y:U:V

In terms of thinking about resolution. 4:2:2 is basically the image has a full resolution black and white image and the colour component is half the resolution as luma. That's for all the colours as they are mixed together in the U and V colour difference signals

4:1:1 - colour is 1/4 the resolution as the luma.

In some formats you have 4:2:0 - that means full resolution luma. Half resolution U for one line and no V. Then half resolution V on the next line but no U = so two lines need to be averaged to get both colour difference signals. So that results in a file that has half the vertical colour resolution of 4:2:2

Basically it's a way of compressing the data without losing much resolution.

It works in the way a water-colour painting works. If you take a splody low resolution watercolour painting and then go over it with black pen and ink - it suddently becomes a high resolution image. The eye responds to the detail in the black lines and even if the colour is soft - the image looks sharp.

In YUV - the fine detail is carried in the luma signal and the colour difference only needs to be good enough. That's why for broadcast 1/4 resolution colour difference is considered acceptable. In YUV - the 3 colours are encoded with the same resolution.

The process stems from NTSC - where they needed to make a colour TV format that was backwards compatible with Black and White TV's. To do that a Luma (black and white) image would need to be transmitted. Its more efficient to transmit Luma + 2 x colour difference. Compared to separate RGB and Luma

Max Field · December 8, 2019

Thanks for that end of it. What about when people say a camera would lose resolution in the red channel? Do different companies select the crushing of the chroma resolution through different methods?

Phil Connolly · December 8, 2019

The formula for conversion is a standard piece of maths , everyone does it the same way:

Y  =      (0.257 * R) + (0.504 * G) + (0.098 * B) + 16

Cr = V =  (0.439 * R) - (0.368 * G) - (0.071 * B) + 128

Cb = U = -(0.148 * R) - (0.291 * G) + (0.439 * B) + 128

Different companies may compress the chroma resolution different ways e.g 3:1:1, 4:2:0. 4:1:1 etc.. but it would affect all the colours together rather than lose resolution in the red's etc... Both U and V contain red information for instance. And you need U and V to be the same resolution to get a proper image...

Historically the chroma resolution would have been crushed using an analogue low pass filter. All low pass filters can potentially create artifacts - so different companies may have used different filters that "might" have produced artifacts that might have caused colours to decode wrong.

NTSC suffers from Phase errors that messed up the proper decoding of colour. PAL flipped the phase of the signal on alternating lines, to cancel out any colour errors - resulting in better colour.

Also with analogue formats the Chroma would need to be encoded on a carrier signal so it can be transmitted - this carrier also can be accientially decoded as "colour" resulting in moire patterns.

In digital land - the RGB to YUV conversion is more accurate and repeatable. Its still a good and simple way to compress the size of the file with limited visual artifact. We don't need Y:u:v any more (since black and white TV's aren't a thing) but it's still ubiquitous in broadcasting and cameras.