[HDR-photo] Following on questions...

[Royce Howland]

John Callan wrote:
> [...]
> Around the 12th exposure I was beginning to get the desired effect and so
> worked on from there.
> 
> Am I doing something horrendously wrong ?

Like Jack Howard stated, there are no hard & fast rules. However 12 - 20 
exposures sounds somewhat extreme to me. I infer from that number that 
your exposures are fairly close to each other, i.e. at 1 EV or less of 
separation.

Doing outdoor shots & stitched panoramics, I get a lot of successful 
images with 3 shots taken at -2, 0 and +2 EV. Other shots require 5 or 6 
exposures, and I have never shot more than 9. In general I use a 2 EV 
step between exposures. How many shots are needed is a function of the 
scene's total dynamic range, how much of that DR must be captured 
(rarely need to see detail on the surface of the sun for example), and 
how widely spaced the exposures are.

In general, fewer shots are easier to post-process, especially to reduce 
the risk of frame-to-frame alignment problems that rob image sharpness 
or introduce ghosts that must be processed out (if you don't want them). 
Most folks I know shoot their sequences at 1.67 or 2 EV apart to balance 
quality of HDR sampling with a reasonable number of images in the sequence.

In my own testing (which I admit is more pragmatic than it is rigorous), 
I have not yet found any compelling benefit to shooting at 1 EV rather 
than 2 EV of separation. This is because each exposure has a "sweet 
spot" several stops wide that contains good contrast & detail; these 
sweet spots are on the upper part of the histogram where the majority of 
luminance levels are recorded. As long as those sweet spots reasonably 
overlap from one exposure to the next, overlapping them even more does 
not buy you any real benefit. (The analogy in another response was 
overlapping shots for stitching.)

The trade-off here is giving the HDR merge algorithm enough luminance 
levels to sample so it can pull in good detail across the entire tonal 
range. Once enough is enough, throwing even more sampling points at the 
HDR tool doesn't help anything. At a 2 EV step I would bet that most 
pixels in "normal" scenes (not heavily biased towards the shadows or 
highlights) have 2 - 3 clean exposures to sample during the merge, at a 
level of detail that is useful enough. But how the HDR tools elect where 
to sample, I'm not sure. I guess the simplest approach would be to pick 
the exposure where the region to be sampled is shifted the furthest 
right on the histogram without being clipped; this would provide the 
largest number of luminance levels to pick from.

Going a bit more than a 2 EV step may help by cutting the number of 
exposures required, although at the risk of pushing the merge sampling 
further apart. But since most cameras have auto-bracket functions that 
are limited to a maximum of 2 EV steps, using more than that means 
touching the camera for every shot to manually adjust exposure. Again 
this risks introduction of alignment errors. Plus it takes a few more 
seconds per shot, which increases the odds of something more significant 
changing in the scene before the sequence can be finished. So I normally 
prefer to use 2 EV, shoot an auto-bracket of 3 shots, and manually 
adjust the exposure to take successive sets of 3 shots if I need more DR 
capture.

This is my interpretation as a pragmatic-oriented guy just trying to get 
the best out of the tools with minimum effort. :) I'd be happy to see 
comments on this from somebody that knows the insides of HDR sampling...

Royce Howland