HDR Techniques

[overexposed]

I'm Looking for some info on the various shooting techniques some of you may have experimented with and some insight into what's useful and what's a waste of time.

1. For example, is the overall depth of field (multiple shot HDR align & merge) increased if the focus point is different for each of the component shots?  The lens focal length remains constant, so the only changes are exposure and focus point.  I tried this with mixed results...

2. Is HDR quality increased if more shots are used in the merge?  ie; 5 instead of 3, 7 instead of 5 etc;  I haven't tried more than 3 and was curious as the affect on quality. Also, how many images (max limit) could one reasonably expect the software to handle before it chokes?

3. How much overexposure would you allow in any particular shot and still consider it useful for HDR?  In many of the HDR scenes I've tried, about 1 stop overexposure (from the determined optimal exposure) appears mostly useless (lots of clipped details).  I assume that all clipped data is simply discarded (no detail) which would mean that each shot should attempt to retain as much detail as possible.


Any thoughts?

[rich_m]

1. Inadvisable - the program isnt really designed for that as far as i know- google "macro focus stacking" and you will find info on how to do that with other programs... maybe you could merge the two in some sort of technique using two programs.  If you are trying to optimise your DOF (esp in landscapes) google "hyperfocal distance" and that will help you.

2. Not entirely sure, but i tend to use 3 for simplicity's sake - one optimal and one either side at 2-stop bracketing - tends to get decent enough results.  Using more tends to complicate matters especially if they arent completely even in exposure range. Some people like to use 5 i hear... but i havent done enough testing / read enough on it to give any real advice except the higher the dynamic range in a scene the more photos/range of exposure you will need...

3.  As above - some of the photos i use for the HDR are clipped massively at both ends, but the whole point is not the clipped parts but the detail in the non-clipped parts (ie the highlights in the underexposed image, and the lowlights in the overexposed image) as that is what the algorithm incorporates into its output.

[overexposed]

Thanks for the reply,
    I am aware of using Hyperfocal distance to maximize DOF, but was simply curious if anyone had tried other methods to maximize DOF and what results they achieved (or didn't). From the multi focal point images I tried, It was difficult to determine if it made things better, worse or had little effect.  I'm just in the experimental stages of HDR and am trying to figure out what works well in a given situation.  In any event, your comments and insights are welcome and appreciated.

[wukong dev team]

Good discussion. I am pretty new with photography so don't believe everything I say. From the engineer's point of view:

1. In the eyes of the image merging software, from hard to easy: Changing aperture alone will result in different DOF on individual images. IMHO it is not a good parameter to adjust if you aim to shoot for HDR, unless you are shooting all landscapes with infinite DOF. Speed is better because DOF does not change; however different speed will introduce different object/camera movement pattern. But that is not all. Changing ISO is another way, but this requires your camera perform well with high ISO (the trend today). Finally, in a studio environment you can change the amount of light, which I think is the best. In real world you should choose one depending on the occasion or combine them in one set.

2. Most people use AEB mode in their camera. If you shoot Jpeg a common rule is +/- 2EV. This means you consider only 6EV of information in each image is "trust-worthy". The other 2EV can be noisy so they are better replaced by data from another image. However, if you camera can shoot good quality 10-bit or 12-bit camera raw files, you can easily make it to be +/-4EV (in 10bit case), +/-6EV(12bit). In theory, the difference in EV among different image should be scene dependent. So say when there is a sun in a window plus a dark room corner plus a lamp could be best taken at -6EV(sun), 0EV(lamp), +3EV(room corner). Usually, if the dynamic range is well covered with say, 3 images, increase that number to 7 will not change the quality much (there is also a downside). Sometimes I shoot 3 pictures at +2EV, 3 pictures at 0EV and 3 at -2EV and select 1 from 3 of the same group then do the merge. The software can handle a very large set (also dependent on the size of each image). I have tried 30 images set. But in the 30 images, more than half of them are useless, either too dark or too bright. They will be actually ignored by the merge so no effect (other than making your CF card full).

3. Exactly, when shooting for HDR you know each image is serving its purpose, that is, a region on the image has proper exposure. As long as these well-exposed regions from every image sum up to be an almost complete image, the software (this or its competitor) should be able to figure something out. So it is a question of what you want the final image to represent. One common misunderstanding is the software can only see what a computer screen can display. No, it is able to extract more information from a 12-bit camera raw file than the jpeg you obtain after processing the raw file (with a regular raw file convector). And a good local tonemapper like DR can bring all that information/detail out in the tonemapping process.

[overexposed]

Nice to see that the development team gets involved in these forums...

Based on your post, would you agree that:  (From the standpoint of what WuKong sees during processing)

1. Using a single focus point is preferable to using different focus points for all component shots used to create an HDR image.
    If true, this would necessarily mean that sharpness does not accumulate appreciably to be of any viable use.
   
    Incidentally, if you select a different focus sensor you can change the point in the image where focus occurs without changing the
    aperture, shutter or ISO and still achieve a change in DOF

2. Using 3 or 5 images in the creation of an HDR image will yield the most detail. (assumes proper exposure)
    Although you have stated WuKong can handle a boatload of images for the merging process, the upper limit would appear to be the
    specific dynamic range for the camera you happen to be using. Typically that would be about 6 or 7 stops with the uppermost and
    the lowest being fairly useless for the purposes of HDR.

3. The highest quality HDR image would be obtained through accurate exposure for each of: Highlights, Shadows, and Midtones.
    Since achieving this would require varying exposure then:
    A) Vary the shutter speed instead of the aperture (DOF considerations)
    B) Keep focus point consistent shot to shot  (maintain DOF)
    C) Vary ISO instead of shutter speed if subject/scene motion is an issue

At least, this is what I've gleaned from what you've said.  Incidentally, I shoot 12bit raw - Adobe RGB: Export tiff from WuKong and do finally adjustments in Silkypix.

One other question...  In the merge process, does wukong randomly pick the image to be used as the base image, or is there some method to your madness?

[wukong dev team]

In the merge process, does wukong randomly pick the image to be used as the base image, or is there some method to your madness?

All images are loaded, examined, and then sorted based on some logic. After that, the one that is most suitable to be used as the base image is selected.

I agree with what you said. Focal length is best left unchanged unless you want to achieve special effects with the tone mapping engine. Refocus on a different point would probably work against you and introduce unexpected results for merge/tonemapping. But I can not say I am the expert on this topic.

[rich_m]

3. The highest quality HDR image would be obtained through accurate exposure for each of: Highlights, Shadows, and Midtones.
    Since achieving this would require varying exposure then:
    A) Vary the shutter speed instead of the aperture (DOF considerations)
    B) Keep focus point consistent shot to shot  (maintain DOF)
    C) Vary ISO instead of shutter speed if subject/scene motion is an issue

A) yes - and most autobracket enabled DSLRs will change the shutter speed only (i think if you are shutter priority it will do otherwise) - if in doubt just use the manual setting to set up your base shot and i doubt the cam will change anything but shutter speed for the bracketing.
B) yup - unless you want to try stacking 3 sets of shots into 3 high DOF photos at different exposures in a stacking program and THEN put them through wukong...
C) is there a camera which can take pictures auto-bracketed by ISO? if not then you are going to have an extremely hard time taking a shot of a moving subject if you manually have to change the ISO... enough that you might as well not bother because each pic is going to be different...

[frate]

The only reliable ways to stack images of moving scenes are best seen in astrophotography.  There are good examples for aligning moving fields (if done properly the field is also tracked by the objective lens) and then using multiple exposures to maintain detail in brighter areas while increasing the relative contrast in darker regions.  Gas clouds are routinely captured this way.  Single pictures of gas clouds are boring.  Astrophotography usually incorporates the near IR as well.

The other approach to moving scenes where the objective does not track is median stacking methods.  In those cases, the object is to remove the moving objects from the final scene and only represent what is most common to all shots, usually the non-moving shots.  The best capture of the moving object can, of course, be blended back in manually.

I'm by no means an expert.  This probably should have started with IMHO.
"What little I know is probably wrong anyway."

- Dean