Color channels clipping independently

[stg]

In digital cameras color channels may clip independently of one another and this leads to color shifts and fringes. Now I understand that this kind of clipping occurs in the demosaicing phase (since RAW data are color "agnostic"). The question is if, by looking at the scene prior to shooting, we can somehow predict that a particular color channel is going to clip. Histograms in LCD screens cannot reveal this kind of clipping, right?

Cheers

[tclune]

Worse than that, histograms and "clipped data" displays on any camera that I know apply to the JPEG data only -- they have nothing to do with the raw information, even if you are shooting raw only. I would really like digital camera manufacturers to pay more attention to raw shooting tools in both their camera and in their toolkits (for example, the Capture NX2 and View NX2 histograms don't actually show the raw data histogram when you are working on raw data -- they show the JPEG version that will result from what you have done. You are basically working with the raw data "in the dark," so to speak. If the tools would show a histogram of the raw data and a box or separate histogram indicating the current JPEG mapping of the raw data, it would be much easier to make intelligent adjustments to the raw files at that stage of the processing. I see no reason that similar tools could not be included with the camera itself. But, for reasons that completely escape me, the camera manufacturers seem to see those idiot JPEGs as the "real" data.

[revi]

In digital cameras color channels may clip independently of one another and this leads to color shifts and fringes. Now I understand that this kind of clipping occurs in the demosaicing phase (since RAW data are color "agnostic"). The question is if, by looking at the scene prior to shooting, we can somehow predict that a particular color channel is going to clip. Histograms in LCD screens cannot reveal this kind of clipping, right?

As colour channels are filtered, yes they can clip independently. This can happen at the moment you take the picture, esp. when there is a strong colour patch, e.g. a poppy. Then the meter (more sensitive to green than to red) can overexpose the red flower. It can also happen in demosaicing when applying white balance correction (basically you scale the three colour channels independently to correct white balance)

LCD histograms can show clipped channels, if the camera shows RGB histograms (not always easy to see though).

Finally, RAW data is not really 'colour agnostic': each RAW pixel has intensity information for one colour, and getting full colour back is the job of the demosaicing algorithm. The information is there, just hidden a bit (otherwise we could never get a colour picture from RAW data). Don't forget that JPG, PNG etc. are also just bytes, the image has to be shown by decoding the information and lighting up points on a screen (or projecting ink drops on paper), the bytes have no colour...

Remco