This is not a definitive article on multi-layer raw processing, just one that introduces a few issues that seem to me significant.
The basic argument has to be - does multi-layer raw processing result in better images.
If you are working with images that have a histogram that look anything like /\__/\ or even worse |\__/| then I think the answer is almost certainly yes. These are images with a lot of bright highlights and a lot of dark shadow and, in the second case, a substantial amount of clipping at both ends.
Here is an example with the image and histogram as they appear under the default raw-processing settings in ACR.
You can see from the histogram that image data is being clipped at both ends. The immediate question on seeing this image is - is the highlight data recoverable? If we have a look at the image with exposure set well below the default position (about -1.5 stops) we find the answer is very much 'yes'.
Unfortunately, when we stop exposure down like that we loose all visibility in the foreground and more distressingly the clouds take on a strong un-natural blue color in dark areas. If we adjust for that cloud color tint using the white-balance dropper in ACR, while the clouds become grey, the trees become unnaturally yellow. So, in the cloud oriented image above I have reduced the saturation by 50% to get the clouds as close as I can to the colors I saw when the image was taken.
I have no doubt that I could produce a better image by using masks to blend the two images above than I could from a single pass of ACR - and I am sure I could improve on the first image to render the foreground how I want it as well.
Nevertheless, let's have a look how far we can go with a single run:
- first shift the exposure setting to -1.5 stops and the shadow setting to 0, thus removing the clipping of highlights and shadows
- second, shift the brightness up significantly to make some of the foreground visible
- third decrease contrast markedly so as to spread out the image data that is bunched at each end.
This is what we get:
Not surprisingly we can see two problems. First, we still have the blue-tinge in the clouds, second the sky is nowhere near as interesting as in the second image above.
That is as far as I am going to go with those images. Processing has not been finished and you could probably achive reasonable results with the last example, but I think it is pretty clear that you could obtain better result from carefully blending and adjusting the first two (and I would aim to improve the first substantially from the default setting).
Raw processing of images is about two things mainly.
First, it is about capturing as much of the image data as possible for later processing. Almost always I will back off from the maximization of the what I get from a multi-layer technique, but I DO like to have the option of doing so from a position of strength - an image with well defined (accurate) contrast and color throughout the dynamic range.
Second, it is about getting the image looking as right as possible first time. Maybe I will tweak the image further with curves and levels but I want to get as close as I can to start with. The less I tweak further the more natural and right my image is likely to end up.
There is a further issue implicit in the 'is it worth it' question. How much more work is involved. My aim with automation is to arrive at a point where I can honestly say it is less work. Less work because I am working with good data that has been carefully processed and blended so the fixing up becomes trivial.
Here is another interesting image. It is a sight I often see on my farm.
The sun has fallen below the hills to the West but is illuminating the sky and the hill to the East. The image you see here is close to it's default raw processing setting and it does not come near to being 'right'. In reality, the bright hillside is a rich deep golden yellow / red/ brown. By contrast the foreground is distinctly blue-green. The reason for this variation is that the image is lit by two completely different lights, one the direct light of the descending sun and the other the post sunset evening shadow. No single white balance is going to render this image correctly - the only way to get it is to processes the areas carefully but independently.
In summary, I think there are two primary reasons to do with dynamic range for multi-layer raw processing - the first to do with making the most of the image information at the top and bottom ends of the histogram, the second to do with variation in white balance in different areas of the image, especially when these coincide (as they often do) with the extreme areas of dynamic range.
Andrew