Tools for better perspective correction in wide-angle panoramas?

Hi,

I'm working on a stitched pano which is 3 images wide - actually even 2, if you drop the overlapping central one. Each individual shot is taken at around 18 mm on a crop sensor, so rather wide. Not unexpectedly the perspective is all screwed up in the stitched picture, but for some reason CS6's Adaptive Wide Angle doesn't seem to work: it underestimates the curvature of the lines.

Here's a small WIP preview of the image:

In somewhat difficult situations like this I also try Microsoft ICE.
Unfortunately it only works on Windows.
It supports Raw files and you can play with the different projection options.
Try the 15 originals and the 3 images.

ICE and other external stitching software wouldn't work. Like I said there's 15 images blended in (selected from a batch of about 60), because the silhouettes are actually brought in from multiple shots over time. PS's Photomerge aligns each source image but masks them without flattening the image, which means I can choose which part to show and mask in or out certain overlapping parts. Since external software outputs only a single layer, I'd have to create one "base" layer from 2 or 3 images, then somehow align all the others on top to blend them in.

Therefore, what I need is something that'll allow me to warp the image back to a proper perspective (it's like a billion pixels wide so loss of sharpness doesn't really matter to me since for all practical purposes I'll be downsampling).

edit: I don't want to sound ungrateful though, and for those panos where I don't need to do any Photoshop witchery with layers ICE does look pretty good. I'll definitely check it out.

armis wrote in post #17706747
edit: I don't want to sound ungrateful though,

Not at all. ICE is just another tool in the digital toolbox.
(and it's free!)

Photoshop Warp tool.

Kirk

Thanks for the heads-up. Didn't know ICE existed. Looks interesting. Downloaded it and will be comparing it to PTGui soon.

So I discovered that if you strip the lens/camera metadata from the file then Adaptive Wide Angle lets you bend the constraint lines to match the actual curve you want straightened, which apparently you can't do when it has recognized the lens/camera combination. Seems like a bit of an oversight to me. Anyway, problem solved.

Aah...is there a reason not to assemble them manually? Looks like a no-brainer.

Panoramas shot with a wide-angle lens (and on a non-panoramic head, to boot) have heavy perspective distortion so aligning them would take not only a lot of nudging, but also skewing and possibly warping. Doing that manually on 15 images is... not something I'd look forward to.

Plus, assembling them wasn't the problem. Even if I did assemble them manually, I'd still have to correct perspective.

armis wrote in post #17708173
So I discovered that if you strip the lens/camera metadata from the file then Adaptive Wide Angle lets you bend the constraint lines to match the actual curve you want straightened, which apparently you can't do when it has recognized the lens/camera combination. Seems like a bit of an oversight to me. Anyway, problem solved.

Interesting, I'll have to keep this in mind. Thanks!

Any chance of seeing the 'corrected' image? I'm curious.

Thanks for sharing!

You did say 2-3 images...not 15...right. Next time, maybe choose a proper lens.

I'm not quite certain what you are trying to accomplish in this image. Is it from 2-3 shots or 15 shots? Did you stitch it manually or use Photoshop or some other panorama stitcher? Are you trying to correct for optical distortion from the lens, perspective distortion from the angle at which the camera is pointing relative to the horizon, or the distortion from the projection mapping that you chose in the stitch?

If you want to straighten the horizontal lines (that is, lines that you know are parallel to the horizon in the real scene) then you can do this in a few ways, depending upon the workflow you choose.

Using Photoshop as the stitcher is a problem when you want to apply the same stitch to different sets of images - you cannot make a template or store the stitch solution so that you can register multiple stitches later on and blend them manually, for example. There are better solutions. Most stitchers will permit you to constrain the stitch with horizontal and vertical lines added to the feature matching points made during the stitch. This tells the stitcher more about the scene and helps level and straighten things, depending upon the FOV of the panorama and the projection you choose.

kirk

The field of view is 3 shots wide. However, there are people in there and they move around. My usual technique is to take a bunch of shots, then combine them in post to keep only the most interesting silhouettes. As a result, while the area I photographed is only 3 shots wide, there are bits and pieces of 15 shots in the panorama.

Photoshop's stitcher aligns all the source images, but keeps them as separate layers. This allows me to use the stitcher on the 15 shots to align them all at once, and then manually adjust the layer masks to show the silhouettes I want ad hide the others. I know of no other stiching program which would let me do that part as easily (which is a relative term, since the blending alone here took me an hour). I'll try to post a screenshot of my layers to clarify.

However, the projection itself is heavily distorted (possibly in part because I picked the wrong one) and that's what I wanted to fix. In previous shots the Adaptive Wide Angle filter had served me well, but this time it seemed to have a wrong read on what a straight line was supposed to be.


@chauncey: I know "buy more stuff" is basically the default response on POTN, but I'm unsure what you mean by "proper lens". If you meant a panoramic head and a T/S lens, well I can only apologize for not having the will to burn cash on lenses I'll use once a year, and for owning a system which, in fact, does not even offer any native T/S lenses in the first place. Also, I don't know what you're talking about since I mentioned the 15 shots right there in the OP, and then again in my second post.


@freestylee: I still have a bit of work but I'll post a scaled-down version when I'm done .

@kirkt: here are the layers, if it helps clarify what I'm doing. The bottom layer is a 3-image panorama (call it a draft, if you will). The rest is a bunch of other images aligned and masked.

I use PTGui, so I can only really comment on your issues from the PTGui toolset point of view. You are basically stitching a 3-shot pano, with multiple images taken for the same segment of the 3-shot composite. If you do not move the camera at all during or between each successive shot at a particular pano segment, then you can perform the initial 3-shot stitch and then swap in other images and regenerate the RENDERING of the pano composite, without restitching. You can also do the initial stitch and then make a template of the stitch, capturing the stitch parameters, etc., and then apply the template to a new set of images to create the exact same stitch from a set of 3 new images.

For example, let's say you took 5 shots at each of 3 pano positions for a total of 15 shots. You look through all of the shots and decide there are elements in all 15 shots that you have to have in your final composite. Presumably, you shot your images by placing the camera at position 1 and shooting images 1-5, then panned the camera to position 2 and shot images 6-10 and then panned to position 3 and shot images 11-15. The result is, that there are no contemporaneous images across all three pano positions, meaning that the images that you combine from positions 1, 2 and 3 to make a pano don't really matter that much.

Choose your best image from each position and then stitch them. Take your time to get the stitch correct, and constrain the stitch with the horizontal and vertical line tools in PTGui if need be. Once you render the stitch and refine it to your liking, including your choice of projection, make your "Master" full res stitch from this pano. Save the stitch as a Template in PTGui.

Now you can bring in your next three images - most likely, the next best three across the pano - and apply the template to them and render the full res version, version A, of this pano. You can layer this pano on top of the Master in PS and reveal the portions of version A that you like, sending it into the Master. Repeat for versions B, C, D so that you will have 5 layered panos in PS that you can blend together. The one thing that will remain consistent is the optical and perspective correction that the stitcher applied and the projection of the resulting pano images. Each pano will be identically registered with the others, so the blending should be easy peasy. If you decide to change projections, you will have to regenerate the panos, but you can save each one as its own "project" in PTGui and just change the projection and re-render as a batch in the batch renderer.

PTGui also has a masking function that permits blending inside PTGui prior to the final rendering. The masking tools may not be robust enough for what you are doing though. Finally, PTGui can render to a flat document, or a layered document - that is, you can render each f the above panos into a PSD that has all of the layering of each segment intact, similar to your layering approach in PS.

This workflow is one way to, hopefully, save you the hassle of trying to align and correct everything on an ad hoc basis and let you get to the "art" side of the composite more efficiently.

Have fun! I like the concept you developed in the thumbnail of the "final" version of your correction.

kirk