Originally Posted by neff2k
Can anyone explain what exactly difference is between the YCbCr 4:4:4, YCbCr 4:2:2, and RGB are? It has been explained when these should be used, but not sure what it really means.
Each signal consists of three channels of information. RGB is simpler to understand. It is a Red, Green, and Blue value -- one of each for each pixel. RGB for devices of interest to us here uses 8 bits per pixel for each of these. Thus you get a 24 bit value representing a mixture of 256 different values for each of R, G, and B.
Now you might assume that Black is a 0 value for each of these and White is a 255 value for each of these, and indeed that's the way it works for "Extended RGB". However when you do video processing of various sorts, the algorithms have trouble near black and near white because the image data has a sharp cutoff right there. There's no "headroom" for the algorithms to take advantage of -- letting things floating around a bit above and below Black and White. And you can get some artifacts because of this.
So video stuff is NOT set up that way. Instead "Studio RGB" uses a value of 16 for each of R, G, and B to represent Black. Reference White is 235 for each. The range from 1 to 15 is the Blacker than Black imaging data -- not intended to be seen, but still in there to make things work better. The range from 236 to 254 is the Peak White data, that IS intended to be seen, but content producers are careful to make sure things still look pretty good for TVs that can't reproduce different shades of white above Reference White. The values of 0 and 255 are reserved.
OK so much for RGB.
Now think of what happens when you get near black, for example. The eye is much more sensitive to shades of gray and to resolution of grays than it is to colors. There aren't that many steps available to keep one or the other color from kind of dominating things, so there is a different way of recording this stuff that works better.
Instead of recording the three primary colors you record a "luminance" signal -- basically the brightness of gray, plus two channels of "color difference" information that says how to color that gray appropriately. In YCbCr the Y signal is the gray scale luminance and the Cb and Cr are the color difference channels.
You will also sometimes encounter YPbPr, which is the same thing, but, properly, represents an analog signal whereas YCbCr is for a digital signal.
Now YCbCr 4:4:4 means that for each pixel you have one sample each for the three channels. Don't ask me to explain what 4:4:4 has to do with this. It is not something man is meant to wot of.
For the devices of interest to us here, each pixel is represented by 8 bits each of Y, Cb, and Cr. And as with Studio RGB, the values for Black and Reference White are 16 and 235 respectively (in the Y signal -- the color difference signals would be at the value that indicates "add no color"). For HDMI V1.3, the "Deep Color" stuff folks are talking about will allow that to extend to 10 bits each or 12 bits each.
Pretty straightforward so far, right? Now it gets funky.
The eye isn't capable of resolving color detail as finely as it can resolve gray scale detail, and back at the dawn of time TV people took advantage of this by only sending color info at half the resolution of gray scale info.
A 480i standard def TV signal has only half as many color samples across each line as it has gray scale samples. It's even worse for analog broadcast signals since color is encoded as kind of a "too rapid" change of gray scale -- which is why you see false colors when someone wears a striped shirt for example.
And YCbCr 4:2:2 is the digital video form of this. It has one color sample for every TWO luminance samples. Like this: Y, Cb,Y,Cr,Y,Cb,Y,Cr.
Now if your connection is able to handle 24 bits per pixel (as is used with RGB or YCbCr 4:4:4) then you can USE this reduction in color resolution to send bigger samples!
Again, your choices are 8 bit, 10 bit, or 12 bit. 8 bit is no fun. It has no better info than YCbCr 4:4:4 and you've ALSO cut the horizontal color resolution in half!
But 10 bit or 12 bit can be pretty nifty. If you send each of the Y, Cb, and Cr channels at 12 bits per sample you still, on the average, send 24 bits per pixel. It fits! But you've got more "dynamic range" for representing gray scale and color scale data.
Well that won't look good will it? You are throwing away half the color! Guess what. It looks just fine because, as stated the eye can't SEE color detail that well.
In fact standard DVD imagery is EVEN WORSE than this! The data on the DVD disc is encoded in what's called YCbCr 4:2:0. And what that means is that you've not only halved the HORIZONTAL color resolution, but you've ALSO halved the VERTICAL color resolution!
Your TV never sees YCbCr 4:2:0. Every DVD player -- even the old, original, non-progressive style players -- "reconstitutes" the vertical color info. I.e., it converts YCbCr 4:2:0 to YCbCr 4:2:2 as part of decoding the data off the disc. It's another kind of "scaling". It is tricky because of differences in video-based and film-based content. Players that screwed this up suffered from the now infamous Color Upsampling Error or "CUE".
And what's worse, the sample size on the DVD is only 8 bits! So you get all the loss of color resolution -- both horizontaly and vertically -- and no gain in data quality. Why do they do this? Because otherwise the movie wouldn't fit on the disc, or it could only be made to fit by cranking up the MPEG2 compression so high that the compression defects would be unignorable.
Next step: Players with digital outputs now have to decide whether to send out that YCbCr 4:2:2 they just created on the fly from the YCbCr 4:2:0 on the disc, and if so whether to leave it at 8 bits or "scale" it up to 10 or 12 bits, *OR* convert it to YCbCr 4:4:4 (8 bit samples) by "reconstituting" the horizontal color resolution as well. Alternatively they could convert it to RGB (8 bit samples) which also requires them to reconstitute the horizontal color resolution.
So if the device you are hooking to the Anthem supports YCbCr 4:2:2 at 10 or 12 bits it *MIGHT* give you better results than YCbCr 4:4:4. Or it might not.
But the only point in doing YCbCr 4:2:2 at 8 bits would be if you believe the source device has a bug in the way it produces YCbCr 4:4:4.
The only way to tell is to try it, calibrating both ways separately, and see for yourself which you like better.