Dmitry Kitsov wrote:Tom, I respectfully disagree. Just because it says 4:4:4 it is not so. I understand that it is important not to confuse what is coming from the sensor or camera processor and a compression scheme.
That being said, debayering algorithms are not magic. They are math. If there is no information in a blue pixel about the green channel, no amount of interpolation will recreate it. While you might have a full RGB data, does not mean it is a valid data.
Here is an example I did for students once. Lets take a canon DSLR and the video it records. The claim is that the video they record is 1920 by 1080. But that is not what being send to the encoder. I'd be pressed to say we get more than half of it is a true data.
I shot the same scene with a 7d and an old canon xl2 3 ccd SD camera. Did an uprez of the xl2 standard def footage to 1080p, and it looked better, and seemed to have better resolution than "natively" 1080p footage coming from DSLR.
And you can take your 4:2:2 footage coming after being encoded as such, from a Bayer pattern sensor and blur your chroma channels in post, et voila you now have full RGB data for each pixel. We are just shifting interpolation to post.
Now I might be still wrong, so if I am please explain, how does debayering recreates the valid data where there was none.
For clarity's sake, I will break down this response to address each section of your reply (in other words, I am not trying to come across as blunt or rude)
1: "
Just because it says 4:4:4 it is not so"
"4:4:4" is a description of chroma sub-sampling - and nothing else. This is not subjective. As the name suggests, sub-sampling is a reductive process, Chroma sub-sampling is a description of reducing the number of chroma samples in a video. It does not refer to any other form of chroma resolution or bayer data or anything else. To describe something as 4:4:4 is to simply say that no chroma sub-sampling has been applied the footage. It is not another way of describing an undersampled image or bayer data or anything else. As a matter of fact - not opinion, non pro-res/DNxHD footage from the BMCC does not have chroma sub-sampling applied, therefore it is true to call it 4:4:4.
This is of course not to say, that the footage has full Raw RGB data for every photosite - it is true that every photosite on the sensor will only contain luminance data for either Red, Green or Blue.
2: "
debayering algorithms are not magic. They are math. If there is no information in a blue pixel about the green channel, no amount of interpolation will recreate it."
Yes, Debayering algorithms cannot calculate the "true*" missing chroma values for each pixel, but they do calculate something. They calculate an interpolated value which, depending on the quality of the algorithm, might be very close to whatever the "true*" value should be. At the end of the process, there IS RGB data for each pixel - whether or not this data is accurate enough or not is not something I am debating - the fact is, there is data there.
3:"
If there is no information in a blue pixel about the green channel, no amount of interpolation will recreate it. While you might have a full RGB data, does not mean it is a valid data. "
The interpolation WILL produce data, as I have said, not exactly the same as a non interpolated RGB pixel, but its still chroma data none the less. The validity of this data can be debated, but the main reason chroma sub-sampling is such an efficient way to save space and bandwidth, is because our own sensitivities to colour is less so than luminance - as such, minute mistakes from the interpolation process are unlikely to be spotted without pixel peeping, and even then, the difference is not massive.
4:"
how does debayering recreates the valid data where there was none."
The key word here is "valid". The validity can be debated, the point I am making, is that there IS data there. There is full chroma data.
Now that these points have been considered, it follows on to compare this to a 4:2:2 video file.
A 4:2:2 video file does NOT have full RGB data per pixel. Accurate or not, the data is simply not there. The image will appear less sharp overall and will have had a uniform process applied whereby chroma data (if it previously existed, from oversampling OR from interpolation), has been discarded.
You can debayer in different ways to produce good or bad chroma results from a bayer sensor - but the result is full RGB data. On the other hand, if you apply chroma sub-sampling to something, the effect is always the same, and you end up with less chroma data - not even less accurate chroma data, but just less data, data where there previously was at least something.
The overall point here is that while the resolution of the Chroma channel can be affected by many different things, the process of Chroma-subsampling should not be confused with nor interchanged with having a bayer pattern sensor and not having RGB data per photosite. There is quite a difference, both practically and technically.
*
True was in quotations because objectively there is no such thing as a true value of an analogue source, they are infinitely divisible and cannot be quantised to an exact value. In this instance I was more referring to what colour value would likely be derived from a different type of sensor which would capture full RGB data at every photosite 
Posted: Sat Aug 17, 2013 11:15 pm
) I'd resist lossy compressed raw, but not lossless compression. 
