Postprocessing, SOOC, and shooting film

[Manfred M]

Good tangential thinking, Tom, but I was not referring to the Bayer CFA pattern.

Indeed, I've never seen it referred to as "4:2:2" either, do you have a link for that?

As you may know, there are two parts to creating a JPEG.

One is compression. The other is degrees of color sub-sampling, including 'none'.

When sub-sampling is used the converter transforms the color space from RGB to Y'CbCr which is a two-axis space. Then Cb and Cr are downsized a) if 4:2:0 was selected but b) only one of them if 4:2:2 was selected and c) neither of them if 4:4:4 was selected.

Point being that the downsizing (called chroma sub-sampling) loses color information which is not regained when the JPEG is decoded back to RGB, e.g. to show on a screen or to get printed.

https://en.wikipedia.org/wiki/Chroma_subsampling

Clear as mud, eh?

And of course going from the 12-bit, 14-bit or 16-bit (some medium format cameras only) data that the cameras capture and outputting it as an 8-bit image.

[dje]

[Tangent]
Is this true? AIUI, Bayer sensors are essentially 4:2:2, with the green channel being the "luminance" value. But I could well be wrong.
[/Tangent]

Hi Tom

I guess there are similarities between the two processes. With the Bayer CFA only one of the RG or G values is captured on any one pixel and the other color values have to be estimated (not perfectly) by the demosaicing process. And yes there are twice as many green channels. With chroma sub-sampling, some chroma information (not RGB) is discarded during the jpeg conversion and has to be reconstructed (not perfectly) during the decoding process. With the Bayer de-mosaicing, this estimation is always done with the full resolution image but with jpeg conversion it could be done with any resolution. I'm not sure how you could tell which process has the most impact on image quality.

Dave