Auto HDR compared with manual HDR editing

[Colin Southern]

I'd gues Collin worked on the darker image. Be interested to know if that's correct?

Hi John,

FWIW, you really need one more "s" in "gues", but one less "l" in "Collin"

I'm not sure which one I used to be honest -- I downloaded all three - opened up the first ... blown sky. Opened up the 2nd (so odds on it being the normal exposure) and it was just right.

The thing a lot of people seem to forget with modern DSLR cameras is "just because they can't see some shadow detail on their monitors" (because monitors will only display around 6 of the typically 12 stops that are captured), "doesn't mean to say that it's not there". I thought that might be the case here when I suspected that HDR wasn't needed for that scene.

As it turned out, all I needed to reveal the detail was a moderate application of the fill light control - and "there it all was" (and appearing very clean as well).

I'm in the middle of shooting about 50 HDR photos at the moment - but only because I want to retain the view through the windows whilst still revealing all the detail of the interior (in other words I'm actually doing it more for the tone mapping side of it than anything else).

More often than not I think people go to a lot of trouble to generate an HDR composite when they can get a faster - AND BETTER - result from a single capture.

[Colin Southern]

Reminds me of a thread a while back ...

HDR vs. ACR

[ajohnw]

It's an interesting subject though Colin. Sorry about the spelling. Some aspects are due to a reduced size keyboard to make space for a tablet. The key pitch is a bit shorter so have to think where my fingers are going without looking.

I suspect the problem with auto hdr in this case is illustrated by the misty shots. I used ufraw to do a straight conversion of each one with no modifications at all and this is what I got.



The centre one has the longest exposure but still doesn't fully recover the detail in the trees and only part of the clouds are blown out. Dangerous to mention with Colin about but I don't think that the tree detail is fully recoverable. Most comes out with a 3 stop exposure increase. That one is ...1167.dng. Colours are flat though and insufficient contrast for a nice shot but they could be real. The one on the right, .....1165.dng looks to be the best candidate for the highlights but may well be too dark to do that well. Not sure where the one on the left would fit in. All suggests to me that the exposures aren't correct for no shadows and good mist/clouds. Hope that's not due to using UFraw. I feel hdr only really needs to be used when there are 10+ stops in the scene and even then it's most likely to only need applying to shadow detail. At that level I don't think there is any need for the usual HDR effects either. As more stops are added the shot gets flatter and flatter so need treatment eg tone mapping.

-

[ajohnw]

I used 2 of the shots to get this. I have lost some of the cloud detail ect as I didn't bother full screening UFraw when adjusting the exposures. The odd effect to the mist on the left seems to be in the shot. I worked on the "tree" exposure and a little on the "cloud/mist" exposure". Both png at that stage. Should have used reduced size tiff really. Saved to 99% jpg and then ran them through Enfuse. No sharpening which may also help.


-

[Geoff F]

There is certainly more tonal range with the misty scene. What I did with the manual HDR effect was to use the highlights image as the base layer then edited in the midtones for the mist and sea. Finally a mix of midtone layer plus shadow layer for the clouds.

Finished with a little extra curves adjustment selectively applied, particularly to dark areas of the foliage.

These shots are the ones where CS5 HDR gave false colours and Enfuse produced dark shadows.

I couldn't locate the Enfuse final image, but maybe my input files were too large.

ps. One extra edit I made was to clone out the telephone wire on the bottom right corner. But we are really only discussing HDR effects here.

[Colin Southern]

It's an interesting subject though Colin. Sorry about the spelling. Some aspects are due to a reduced size keyboard to make space for a tablet. The key pitch is a bit shorter so have to think where my fingers are going without looking.

I suspect the problem with auto hdr in this case is illustrated by the misty shots. I used ufraw to do a straight conversion of each one with no modifications at all and this is what I got.



The centre one has the longest exposure but still doesn't fully recover the detail in the trees and only part of the clouds are blown out. Dangerous to mention with Colin about but I don't think that the tree detail is fully recoverable. Most comes out with a 3 stop exposure increase. That one is ...1167.dng. Colours are flat though and insufficient contrast for a nice shot but they could be real. The one on the right, .....1165.dng looks to be the best candidate for the highlights but may well be too dark to do that well. Not sure where the one on the left would fit in. All suggests to me that the exposures aren't correct for no shadows and good mist/clouds. Hope that's not due to using UFraw. I feel hdr only really needs to be used when there are 10+ stops in the scene and even then it's most likely to only need applying to shadow detail. At that level I don't think there is any need for the usual HDR effects either. As more stops are added the shot gets flatter and flatter so need treatment eg tone mapping.

-

Hi John,

I'm confused ... you refer to a DNG file, but I can't see a link to download any of them. Are they available somewhere?

[ajohnw]

Me too Geoff. I had another go and this time did the clouds properly. That funny mist comes from the tree exposure and so far have found no way of getting rid of it.. If you have a problem with the enfuse gui you are using try contacting the author. They usually will fix bugs or help. It's interesting to play around with files like this. I'm getting the impression it's best to process each image for what is wanted in the final one and then merge them. This is enfuse again.



Basically I used the raw tone curve to adjust the wanted bit and then distorted the curve to obliterate as much of the unwanted as possible but keeping the wanted bits part of the curve linear at the default rate. This needs a new language. I then tone mapped and adjusted each image after removing some residual not completely black from the high light shot.

Must tell Colin it's fun even if not needed.
-

[ajohnw]

They are back in the early post Colin. Some way down the 1st page - mediafire links.

I have fiddled with the odd mist problem on the left of the last shot. Not tooooo well though.
-

[revi]

To be precise, the links are in post #9 of this thread

[Colin Southern]

Here you go Geoff ...

[Geoff F]

John, those mist edges are a tricky area, both left and right sides. They seem to have just come out correct on Colin's edit, and my manual merge, but your software has shown it as a more solid area.

The CS5 merge also gave problems with false colours in the same mist edges.

When using the CS5 HDR method I have previously given the images a little tweak during conversion. But the CS5 advice mentioned using direct Raw files for the merge, so that is what I tried. I don't think it came out so well though.

The warning probably refers to anybody doing excessive alteration during conversion.

All the methods seem to be recovering detail from the dark foliage areas now.

[ajohnw]

How was that one done Colin?

I had another go at the left mist. There is a slight yellow cast there in an over exposed shot. Removing it didn't improve things. I think the problem is that the 2 tones come out equal so the merge code doesn't do the correct thing. Latest take - copy, paste,blend and blur over it. I need more practice at that but it's wouldn't be too bad if I didn't know it has been done that way


-

[Colin Southern]

How was that one done Colin?

I took a look at just using one file, but the shadows were just a touch noisy (close though), so I just used Photoshop's Merge to HDR Pro, and hand-tweaked from there. Usually the program squeezes all the data in, but doesn't always do a great job with local contrast - so I just tweak that manually.

[ajohnw]

I have seen instructions on the web suggesting the "correct" exposure and +1 and -1 stop. I haven't gone out and taken any shots strictly for HDR just awkward exposures to see what I can get out of raw files that way but I don't think an always do this will will work with multiple exposure hdr shots. There is a need to think. Taking the mist shots for instance. A slower exposure would bring out more detail in the trees and the land to the left. It would also blow the highlights out more making it easier for the software to work. More often than not I suspect +/- 1 stop wont achieve anything as it didn't on the creek shot as the middle exposure was correct. If it wasn't one of the other 2 would probably do the job.

I have also seen comments in camera reviews in particular that say 3 shots is "better". That so far looks to be a load of crap to me. It depends how much tonal range needs to be covered.

The web is a great place but this reminds me of the "great shoot to the right debate" where shots illustrating it contain some significant white and the exposure comes out right. now isn't that surprising.

It's nice to have some one else's shot's to play with as the UK weather is crap at the moment. If I watch carefully there has been the occasional hour of decent light. Blink and it would be missed. I hope to try using all 3 shots sometime today.
-

[Colin Southern]

I have seen instructions on the web suggesting the "correct" exposure and +1 and -1 stop.

In short, it's bad advice -- for two reasons

1. The difference between the noise levels in any 2 stops is approx nothing, and

2. Going +&- 1 stop extends the dynamic range by 2 stops, but the "correct" amount is totally dictated by the dynamic range of the scene being captured (there is no "one size fits all).

[ajohnw]

I would have thought that the simple answer would be to gauge results using the camera preview Colin maybe trying to spot read medium tone values at the various levels needed. I'm not a fan of spot metering, a smallish area is better but ideas on that score vary from camera to camera.

This is an interesting area on a Pen. The electric view attempts to mimic how the shot will turn out. If that fails over and under exposure are directly indicated in the preview. That can be seen via the eyepiece if the camera has one. Maybe other cameras do the same thing or something similar. If the wanted tone level is ok/perfect in preview it should be ok for merging with others.

-

[Geoff F]

One thing which I have tried when shooting a bracketed sequence is to push the exposure bias in one direction or the other, with standard exposure compensation, as well as the 3 shot extension.

When looking at a tricky scene, it is often obvious that problems will occur in either the shadows or highlights. So using a bit of exposure compensation (say +1) will give an effect similar to 3 shots which range from 0 to +2 for example if using one stop increments.

[Colin Southern]

I would have thought that the simple answer would be to gauge results using the camera preview Colin maybe trying to spot read medium tone values at the various levels needed.

No, not at all - I usually just stick with evaluative metering and then get the camera to bracket 3 to 7 shots from there.