Photo Merge (Perspective) Applies Unwanted Perspective

[Blaine Beron-Rawdon]

   I have had good results when photographing flat art or a full roll of negatives (contact sheet) by making multiple exposures in the form of a panorama. The panorama is then merged, increasing resolution. I use the "perspective" setting for the marge in the preview window. 

   I have trouble on occasion when the artwork or contact sheet contains, for example, buildings. Lightroom may apply a correction that is wildly erroneous. The "cylindrical" and "spherical" settings do not work at all.

   Is there some way to suppress this perspective "correction" and simply merge the photos?

   Four example files are uploaded here. 25, 26 and 27 are three shots of the same two-and-a-quarter contact sheet, taken in pieces. 25-2-Pano is the panorama created from these three photos. Clearly, this is not what I had in mind! I have used this process on many rolls with less architectural subjects and it works perfectly.

   My hope is that I am missing a setting and that this problem can be eliminated. If not, it would be great if Adobe could fix this issue. 

   Thank you for your attention.

[Conrad C]

I don’t think there is a way to change how Lightroom Classic merges panoramas, other than the options you see. As far as I know, all of the projection options are going to assume it is a real world scene and look for perspective clues that it can use to build the panorama with the distortion expected of that projection type.

You may have gotten lucky on the ones that work, if Lightroom Classic paid more attention to the scan grid and that caused it to think it was like a flat wall shot head on and should not be distorted. The only suggestion I have for Lightroom Classic is to maybe try arranging the same negatives differently in the scanner, because for example, I notice that the negatives that got distorted also happen to have light leaks (the black edges) that the others don’t have. I wonder if those blackened edges cause the application to alter how the application perceives the grid between the scans, which is almost white for the other scans. I don’t really know how much it would help, but for example I would try putting all blackened negative edges at the outer edge of the arrangement.

Right now, within Adobe photo software, the only way to merge them while preventing distortion is to do the merge in Photoshop instead. Because in Photoshop, the File > Automate > Photomerge command (or Tools > Photoshop > Photomerge in Adobe Bridge) is not only built on different (probably older) code, it also has different options than Lightroom Classic. For a job like this you should try the Reposition or Collage option. Neither distorts; Reposition is only allowed to move the images to merge them, while Collage is also allowed to scale or rotate them.

Of course what comes out of Photoshop Photomerge is not a .DNG file but a Photoshop document, in case that matters.

If you want Lightroom Classic (or Adobe Camera Raw) to add a Reposition option, it would be best to submit it as a feature request in the Ideas subforum here.

[richardplondon]

I tried your same 3 images in another stitch program, and that also failed to detect the alignments properly. So the default stitch became a perspective distort even though that program does normally understand and detect 'translational' stitching -  a stitch circumstance that LrC does not directly address, I would guess since this is not a classic photographic type of situation.

These images did stitch well once aligned manually using explicit control points - which is more of a specialist stitching technique and IMO outside the intended scope of both LrC and vanilla Photoshop - so I think we can point the finger at how the picture content is getting interpreted within these software applications, rather than at something genuinely technical about the images. So you have both varying framing elements which are not on exact rectangular grids, and inside those frames, varying represented perspectives including some double exposure where mismatched perspectives coincide. I needed to (in effect) tell the software to disregard everything that it "saw", and operate blindly by specific instructions.

With such multiply-contradictory hints to follow, it is not surprising to me that an automated algorithm which looks for a single common perspective "world" within which to find and allocate the picture geometry, could get badly confused!

[Blaine Beron-Rawdon]

Thank you to Conrad C and richardplondon for your thoughtful and thorough replies. I had a thought in the middle of the night to attempt a merge of just two of the three photos, and then, if that worked, to merge the third photo. Remarkably, that worked perfectly as shown in the attached "IMG_0025_2-Pano-2-Pano.jpg". So my immediate problem is resolved with a workaround. Thinking more about the problem, I think there might be a programming solution. A common use of "perspective" merge is to enable a single planar projection of multiple photos taken at different orientation angles. In my case with flat art, those different angles are almost identical. Perhaps Adobe could add a fourth projection type, perhaps "Flat", that limits the merge seek algorithm to a small range of angles. This would prevent wild misinterpretations of flat art.

     An additional point: simply sliding the images around until they line up will not work to the standard needed for this task. In particular, even minor lens distortion (barrel or pincushion) will result in undesired image mismatches at the edges. Lightroom's accommodation of distortion is remarkable.

     As suggested by Conrad C, I will submit my request through the official "Ideas" portal. Thank you agin for your help - it is much appreciated. By the way, these photos were made by my brother Matthew about 50 years ago in Germany when he was a high school exchange student. He used an ancient Rollei twin lens reflex with uncoated lenses. Scanned indivdually, I think some of the negatives may clean up pretty well.

[richardplondon]

There is a fundamental difference between what I described as "translational" stitching and the presented choices - perspective, cylindrical and spherical - since the latter three all presume rotating the camera around a fixed common eyepoint, and then mapping the constructed view that results, as one or another output projection.

So yes, what you are doing does IMO call for a separate stitching option that invokes a different algorithm behind the scenes. OR - for purposes of this mosaic preview - you could consider taking the component shots by rotation rather than translation.

3 shots, one head on, one rotated a little one way, one rotated a little the other way, using some nature of pano bracket (which places the axis of tilt on the lens axis and in a forward position that coincides with the lens 'entrance pupil' - its own natural eyepoint). If you don't have a manufactured bracket, nodal slide or focusing rail or whatever, something would be relatively easy to engineer from scrap stuff (much simpler to do for just one body and lens, and only one axis of rotation). The resulting shots would then be the exact type of input that LrC's pano-stitch is already designed to receive and recognise properly. Selecting Perspective as your output option. 

Alternatively the PS route that has already been suggested.

[Conrad C]

…I had a thought in the middle of the night to attempt a merge of just two of the three photos, and then, if that worked, to merge the third photo. Remarkably, that worked perfectly…

By @Blaine Beron-Rawdon

Good idea. We should have thought of that, because I’ve used that incremental merge workaround for normal (real world) panoramas where for some reason Lightroom Classic could not merge an entire set at once.

I saw your feature request and upvoted it, because even though I don’t need to do exactly what you do, I have used the Collage or Reposition projections in Photoshop to merge oversized art scanned in multiple pieces. So I wouldn’t mind having a similar “Flat” type of projection in Lightroom Classic.

[Blaine Beron-Rawdon]

Conrad C, thanks! 

[Jao vdL]

I see you have gotten some good answers already but I am curious. Is the intent to create a higher resolution image of the whole contact sheet? You can't create higher resolution scans of the individual images in the contact sheet this way looking at the indivudal shots you're trying to merge as the individual scans are bigger than a single negative.

[Blaine Beron-Rawdon]

Jao vdL

     I am making contact prints of an entire roll of film, usually 35 mm. The idea is to make a contact print that is large enough to view in some detail on the computer monitor. I am doing this by arranging the negatives in six rows of six images which makes a sheet of about 8 x 10 inches (200 x 250 mm). This horizontally-oriented sheet is rotated 90 degrees so the wide dimension of the camera spans the 200 mm dimension of the sheet. I make three shots - top, middle and bottom of the now-vertical sheet using a camera mounted on a modified enlarger. These three images are merged in Lightroom to make an image that is about 13,500 x 8700 pixels in size. (I am using a 50 megapixel camera.) This results in individual 35 mm frames that are about 270 x 406 mm on my monitor.  I think this is very useful. 

     After selecting individual frames, I photograph each negative individually, yielding an image of about 45 megapixels that can easily resolve film grain. Prints from these scans look good. I am quite satisfied with this method.

     Please let me know if I have failed to address your point.

[Felicity28029150ikaq]

Hello, 

Sorry to jump on this, but does anyone know if there is a way to calculate the horizontal field of view from a stitched panoramic image? 

[richardplondon]

All sorts of clever stuff is possible in stitch software that is based on the open source PT-Tools suite. The measuring of  field of view must be somewhat non-linear for partial panos, since various different output projections / mappings are used and often in a 'black-box' manner.

I'd only observe, we can at least be certain with a 360 x 180 degree full spherical pano image in particular, that (e.g.) one third of the width of the whole thing must represent 120 degrees out of that 360. 

But the only numerical readouts I have personally seen in the programs I've used (and there is nothing like this in LrC or Photoshop as far as I can see), were more about the attributes of the component shots, rather than about the entire composite.