Understanding Dynamic Range

This article is a work in progress, as there is plenty more to say, but since this is a frequent topic in our forums and others I thought I'd start by reposting my reply to a question on the colortheory mailing list:

>
> But in a raw file, haven't you got about another two stops at either
> end that are recoverable using the exposure slider, on top of the
> standard exposure?
>

The short answer is "it depends". I did, along with Stanford's imaging lab (and published) a test very similar to the Clark's test you cited in our eBook using the D1X (it was a couple years ago when it was current). We found the same 10-stop total limit--interestingly 10-stops was the limit of the lens as well, as lens flare limits the average pro zoom to a 10-stop range. FWIW, even though Clark's target has a 10-stop range, if you look at his transfer function it covers at most 9-stops even for the D-SLR.

BUT, much like the test image on Clark's site, our test target had nice solid patches, and the bottom 2 stops were noisy and the top 2 stops were measured numerically by pushing the exposure up until white just saturated the sensor at around 4095. Neither is very realistic compared to trying to photograph real scenes at those extremes.

So yes, if you have to salvage highlights or shadows there is some information there you can get to, but the dark shadows will be ugly with noise. And advanced Highlight Recovery tools are actually recreating the burnt out channels (which may be > 4096) by synthesizing them from the channels which are still holding information. A very cool technique, but not exactly high-fidelity image capture.

[Both of those points are sort of important. Test targets tend to have Black and White at the bottom and top, for obvious reasons. But your camera has different amps for each channel, and one will always burn out first, or go dark first. So unless you're shooting test target with perfect white balance you're losing information at the top and bottom before your pixels completely blow out. Advanced Highlight recovery tools can recreate some of that data, but they are doing it by estimating what the blown out channels might have been, not because they find a secret stash of information.]

Net net, if you go back to the white bird against grass or a bride with a white dress standing next to a groom in a black tux I've never seen a case where you can actually _use_ 10-stops with a D-SLR and hold the detail at both ends. You can prove there are 10-stops in the image, but (at least in my experience) you can't use them

So on a test target sure 9-stops, 10-stops, why not. But, like the highway mileage cited for many cars, useable dynamic range for the photographer isn't the same thing. With the first generation D-SLRs I used 4-stops as my useable range. With my D2X I'd say 6-stops are normally quite useable, maybe even 7 depending on the scene.

FWIW, I lead a lot of field workshops with lots of Nikon & Canon shooters, and the 5-6 stop guidance to participants works out great. Without exception if they try to capture scenes with more than that in a single frame they wind up with a long and usually disappointing Photoshop session [even with Raw files & 16-bit processing] For more range than that if the shot is a scenic then we use multiple frames & combine them. If it is wildlife then we tell them to pick whether they want to preserve the shadows or the highlights and let the other one go.--