Which is easier to recover from a raw image, shadows or highlights ?

Question

Obviously, severe under or over exposure is not recoverable but in moderate cases ?

Answer 1

Slight overexposure is generally preferable to slight underexposure because of the way the raw images are stored. Raw generally has 4096 discrete values (for 12-bit raw) in each channel (red, blue, green) which are then mapped to 255 values (for 8-bit images) that you see on screen.

In raw, half of the information available (2048 values in each channel) are taken up by the brightest stop. The next brightest stop takes up 1/2 of that (1024), the third brightest, half that (512) and so on. What this means practically is that highlights have a lot more data available to them than shadows.

As long as you don't blow any of the channels in raw (meaning exceed the brightest 4096 value), you can recover bright details by remapping the pixels. Unfortunately, there's no way to tell exactly how much headroom your image has on the camera, as the histograms and highlight warnings apply to the JPEG conversion of the image, not the raw sensor data. I generally find that you usually have about a stop to almost 2 stops of headroom when the highlight warning starts to blink on your preview on the back of the camera.

So, what happens if you try to recover shadows is that you're using a much smaller set of values (because the darkest stop has around 64 levels), is you're trying to spread the 64 values over a larger area, which can cause shadow noise and posterization. When recovering highlights, you're compressing 2048 values down, so you're not losing any information.

However, once a channel clips at 0 or 4096 in the raw data, there's no meaningful detail you can extract from areas darker or brighter, respectively, in the picture.

The general rule is, with raw, slight "overexposure" is preferable (look up "ettr" or "expose to the right") and with JPEG shooting, underexposure is preferable.

This is a rather involved subject, so I'll give you a few resources to read:

http://www.ttb.gov/faqs/genalcohol.shtml...

http://www.luminous-landscape.com/tutori...

Answer 2

In my own (limited) experience, I have found shadows easier to recover. Noise reduction can be quite helpful, and the underexposed image still has all the right structures in the right places, whereas in an overexposure often has colours or value bleeding into neighbouring pixels which ruins the actual shapes in the image.

Also, since under- and over-exposure relates to exposure durations, it is helpful that under-exposures are shorter (and thus have less shake in them).

Answer 3

A given range of the deepest shadows will have only about 1/20th the information of that same range of the brightest highlights. That means that, no matter how careful you open shadows, you're going to introduce some noise. Since, in most photos, shadow detail is as important, if not more important, that highlight detail, it makes sense to keep the shadow detail from clipping and try and recover highlight detail.

Honestly, though, how many scenes have such a dynamic range that we are forced to choose between one or the other? In the majority of cases, the scenic tonal range will fit nicely within the range handled by the camera, especially of shooting RAW. Sure, some scenes have a tonal range that exceeds that which can be handled by the camera's sensor, but those instances will have to be dealt with on a case by case basis, depending on the important areas of detail within the image.

But as an overall sort of answer, recovering data from highlights will yield much better results than doing the same with shadows.

Here's a good discussion regarding information contained within highlight and shadow areas.

http://www.luminous-landscape.com/tutori...

p.s. Sorry Peter, didn't meant to copy your links.

Answer 4

I find shadows easier to deal with. Digital cameras seem to burn out the highlights really easily. I would rather underexpose and bring up the shadows than to overexpose and try to repair the highlights.

Part of the problem is that highlights tend to be small and have borders that show changes in the image very readily. When I feather my highlight selection to help smooth my modifications, I can't use a very large feather value because the highlights themselves are small. With a small feather value, you have to be very careful or the edges where you made the modifications will be easily visible.  

With shadows,  you are often working with larger areas and you can feather the changes better. Small areas of shadow are rarely important.