I did learn something about my camera by participating in this thread, and began to think how I would use its features to implement HDR. My main camera (Nikon D3) allows AEB with up to 9 shots, but only at a maximum of 1 EV steps. If only 3 shots are needed one can set the camera to take shots at the nominal exposure, +1 EV, and +2EV. If 5 or more exposures are set, it brackets up and down around the nominal exposure, but one can use exposure compensation to counteract that so it effectively brackets only above the nominal exposure.
Hi Bill,
That's the usual way of doing it, bracketing with EV compensation to get a bias towards longer exposure times. Otherwise there will be too many exposures covering the highlights, where we actually need more photons to improve the shadows. The usual shortcomings of various cameras in facilitating an ETTR exposure determination based on Raw data keeps hunting us though.
[...]
Alternatively, one could make manual increments of 2 or 3 EV above the base exposure, but this would require touching the camera. One could use autoalign to take care of camera movement. In our host's Camera to Print and Screen tutorial, he states that he often shoots HDR hand held with good results using auto-align, so touching the camera on a stable tripod might not be all that bad.
There is something else that needs to be considered, and Guillermo and I probably have a different opinion about it (because we use different software), the large bracket steps. In my experience, using various HDR assembly applications (for building an HDRI from bracketed exposures), the larger exposure bracket steps can cause various quality issues. To name an obvious one, image alignment. When the individual exposures are several stops apart, the noise profile at the transition zones between the exposures may prohibit finding a good alignment automatically. Even shots taken from a tripod may benefit from a (sub-)pixel alignment. Another one would be the possibility to automatically address ghosting artifacts through pixel variance.
Recent tests I have done with SNS-HDR, which is basically an (unparalleled) exposure fusion application, show me that I can push the bracketing interval to 2 EV steps, but then beyond that inevitably I can see increases of noise occuring on smooth gradients (of course depending on the amount of postprocessing). The noise increase can probably be measured already at smaller intervals, but I focused on visual artifacts. The author of SNS-HDR, who knows what is going on under-the-hood, suggests to limit the intervals to 1 EV for the best results. I often use 1.33 EV intervals when I have to cover a large scene dynamic range in a somewhat limited number of exposures.
Other applications, notably the HDR assembly and camera response curve reconstruction types, can also suffer from large exposure increments to varying degrees.
Gregory Ward Larson, one of the pioneers of HDR imaging also suggests to use 1 EV bracketing steps for the assembly of HDRI's. That doesn't mean that we shouldn't think about scenarios where larger steps are useful, but there are often compromises involved when we do.
Guillermo's ZeroNoise program is probably unique in the sense that it exploits the large EV step difference, but it also assumes minimal ghosting and perfect image alignment. I am slightly concerned about (shot) noise differences at the exposure transition zones. With e.g. 4 EV exposure differences, the noise varies by a factor of 2 (although it will be reduced some by gamma conversion), which could easily be picked up by the ensuing tonemapping operations, especially unwelcome on smooth gradients.
Some of the noise risk is minimized by shooting at low ISO's and by using the ETTR zones of the brackets), but that applies to all methods that will require significant post-processing.
Cheers,
Bart