HDR Merge Windows

[Peter Michael]

Affinity Photo 2.1.1 Windows

Merged 5 jpg files  _DSC8701 - -DSC8705, 1 stop exposure variation in each.

Options selected below.

Selected Dramatic 

 

Resulting image - Merged_8701_8705_ArtefactsHighlighted  is that there is some ghosting & some artefacts (highlighted in red boxes).

I note that this issue has existed for a long time - likely since the release of V2.

 

 

 

[winfried_z]

Peter Michael:

As a general remark, I have often observed that the option "Automatically remove ghosts" creates artefacts. I think this is worth to be considered by the developers. As a workaround, it sometimes helps to generate two versions of the HDR-merged image, with this option on and off, and eventually mask out the artefacts.

Concerning your pictures: The person's position is so different in the individual pictures of the series, that it is very problematic, if not impossible, to create a good alignment.

My advice in this specific case: I think it is not necessary to merge the pictures. I suppose the pictures with nominal exposure or +1 stop include enough details in both shadows and highlights, so you can apply a tonemapping on the single pictures, if this is desired.

General advice: Shoot in RAW, develop to 16 bit TIF, then you have much more information in the pictures than with JPG.

Winfried

[Peter Michael]

Thank you very for your thoughts Winifred.

I have been able to incorporate HDR into much of my photography in a similar situation to that which I have described - taking 5 images with 1 stop apart.

My previous use of infinity photo for HDR or prior to that when I used Photomatix - specialist HDR software, I did not experience ghosting or artifacts even for quick moving subjects. I just tried using Affinity Photo 1.10.6.1665 and found that the same issues - ghosting and artifacts were evident.  Accordingly, I suspect that the issues started occurring some time with Affinity Photo Version 1.

Peter

 

[NotMyFault]

In such situations with moving people you get better results by manually de-ghosting with help of clone tool / sources panel.

The automatic de-ghosting has no AI to e.g. identify people, so it can select random parts of people only based on „best lighting“ which is not suitable.

This is more a principle limitation of the used algorithms, not a bug.

[Paul Mudditt]

The Affinity Workbook example shows you how to correctly removes ghosts.

[winfried_z]

Peter:

I think the the problem here is not the quick but the slow motion of the person, causing partial overlaps in all five pictures.

In addition to what NotMyFault suggested, If you wish to fix your final picture (the one with the red boxes), you could simply copy a rectangle of the +1 stop picture containing the person over it and and adjust this a little via clarity, tonemapping, brightness, ... to obtain a seamless transition, just like that:

 

NotMyFault:

In this case, I agree. but I found much less challenging cases where artefacts occur, which I definitely would call a bug, e.g. here (3 TIFFs, 16bit, -2/0/+2 stops, "Automatically remove ghosts" on):

Winfried

[Dan C]

24 minutes ago, winfried_z said:

In this case, I agree. but I found much less challenging cases where artefacts occur, which I definitely would call a bug, e.g. here (3 TIFFs, 16bit, -2/0/+2 stops, "Automatically remove ghosts" on):

 

Are you able to provide the source TIFFs used for this merge, so I can test this issue further and report it to our developers please?

[NotMyFault]

37 minutes ago, winfried_z said:

Peter:

I think the the problem here is not the quick but the slow motion of the person, causing partial overlaps in all five pictures.

In addition to what NotMyFault suggested, If you wish to fix your final picture (the one with the red boxes), you could simply copy a rectangle of the +1 stop picture containing the person over it and and adjust this a little via clarity, tonemapping, brightness, ... to obtain a seamless transition, just like that:

 

NotMyFault:

In this case, I agree. but I found much less challenging cases where artefacts occur, which I definitely would call a bug, e.g. here (3 TIFFs, 16bit, -2/0/+2 stops, "Automatically remove ghosts" on):

Winfried

I would be interested getting the source images, too.

using +/- 2 stops is on the extreme side for HDR merge. I would not go above 1 or 1 1/3 stops, otherwise use more images (5 or 7). 
I doubt that the scene you are showing here has such an excessive dynamic range requiring +4 stops. A normal APS C DSLR covers 11-13 stops, a FF up to 15 in a single RAW. 

HDR stack requires careful planning. Most of us (including me) take far too many images covering a much to wide theoretical DR. Using a current DSLR you can get HDR look from single images, only backlit scenes may require more - but this somehow negates the look of backlit images.

[winfried_z]

Dan:

Here come the source TIFFs, and these are the settings:

For demonstration, I chose a high local contrast and reduced the brightness, but you can see the effect already in the merged picture without tonemapping.

NotMyFault:

The centre picture has already an exposure bias of -2/3 stops; still the sky is overexposed. And further reducing the exposure would create additional noise in the shadows. Maybe +/-1 stops would have been enough, but in any case, the artefacts should not be there.

Winfried

DSC_3560.TIF DSC_3561.TIF DSC_3562.TIF

[Dan C]

Many thanks for providing those for me! I can confirm that I have been able to replicate this issue here and therefore I have logged this as a bug with our developers now.

I hope this helps

[NotMyFault]

ok, issue reproducible for me.

The image differ extremely in noise profile. It seems it uses mainly 3561, which is the best source wrt to colors an overall brightness in general, but for some areas in sky, it uses 3562 which has no noise and is actually almost outblown. The tone map persona makes this more obvious with high local contrast, the damage was done during stacking and closing wrong source areas, or wrongly match colors and brightness.

When I try to manually clone in from 3562, Affinity is unable to match brightness, and produces a much to dark color.

This is another issue, unrelated to ghost removal.

 

[NotMyFault]

I've tested with Photo V1, getting identical results. so not a new bug

[winfried_z]

45 minutes ago, NotMyFault said:

This is another issue, unrelated to ghost removal.

I don't know what the root for this behaviour is, but when I merge the three images without de-ghosting, the result looks neat:

And as another point, if only 3560 and 3561 are used as sources, but with de-ghosting, the result is very similar to the one using all three pictures with de-ghosting switched on; thus, 3562 cannot be the villain:

Winfried