Thanks so much for the lighthouse example...the reason I havent been working on this is I don't have a digital camera to take some pictures with so including the files is a godsend for me.
I am guessing that the repeating pattern in the sky is one of two things, uneven illumination from the light source or vignetting from the lens. In either case one solution is to take a picture of the light source without the negative on it and use its negative as a layer on each picture before merging. Just a thought. I guess I am assuming that the negative moved around a stationary light source instead of both picture and light source moving together.
I am excited to start working on merging these tonight after work.
I've done a little testing, and I'm confident that my light source is the issue. It's not fixed down, and putting the flash in position can very easily cause the the whole source to slide. This gives a dark pattern very similar to what I see in the scan. I'll built a better light source soon. When I do, I'll upload more files for people to play with.
Hi Peter, I have some remarks. First, the shots are not that sharp. I am not sure in that, but it can be. Second, autostitching software can be confused with dark areas around the shots. It is better to shot without them. Looking forward for new files. Al
A quickie test using Jim's source images. I simply ran these through Autopano Pro on my default settings. On my first try I had the strong banding/shadows as in Jim's test shot. Then I cropped the black areas out and tried again. Same result. Autopano didn't seem to care about the black areas. That was surprising. Then I vignetted a little to compensate in a rough guess sort of way for light fall-off in the lens. Stupidly I did this to the cropped images so not the best test (it's past midnight here, sigh). In any event, this seemed to lead to a noticeable improvement in the shadowy banding issue.
Autopano refuses to even add images to the group if it cannot find control points to align them with, thus the voids. Does anyone know a way to manually insert and roughly position featureless images into place within Autopano? PTGui does this easily. For all of Autopano's power I suspect I'm simply overlooking that option.
As we know, this module of our project is focused on stitching. This, I expect, is the biggest hurdle we are faced with. If we can solve this, then we can worry about different lenses, focus stacking, various types of light sources... But it doesn't really pay to spend time on those issues if we can't solve this. I really appreciate all of the help I've received on this, as I'm a stitching novice.
With photos with details everywhere we've already had success, but the problem is images with large, homogenous and non-detailed areas, like those in the light house. It is not a good photograph, and I expect it isn't very sharp, as it was super windy that day, and I used a 10-stop neutral density filter. But for our current purposes, it seems like a good choice.
So far, I've had the best success with PTGui's align to grid feature, with the recurrent pattern in the sky being the main problem. This is most likely due to a light source issue. I'm almost done building a new light source, and it should be considerably more even. In the future I'll also include a picture of the light source, which might be usable to get rid of the variations from the pictures. I should have the new source down by mid-day, and when it's done, I'll shoot a new series of shots.
In addition to PTGui, I"ve dabbled a bit with Autopano Giga. Here are the results so far:
Unfortunately, Giga doesn't allow me to use their align to grid process, as that seems limited to mosaics shot with the various motorized pano heads.
Would using something like Adobe® Lens Profile Creator be useful? With it we could minimize geometric and tonal distortions introduce by our lens before we stitch the individual frames. On this topic, does changing magnification change the distortion characteristics of a lens? The ALPC uses photographs of charts to make it's profile, and these would be easiest to make and photograph at non-macro magnifications.
I'm hoping that the very precise and repeatable negative stage movement offered by a motorized x-y stage will help with the stitching issue.
I've used Microsoft ICE(free BTW) on my 5DMkII stitches w/ very good success.
However, I have not tried it on stitches that have large expanses of openness, sky, water, etc...
I don't have any, have to generate some this weekend.
There is also another freeware stitching app, Hugin, I've used it, but it demands much more time and knowledge.
MS ICE was sooo much easier.
PS CS5 was so so to me.
Also, my experimenting of stitches were not related to this current project, I just learned of this project recently.
Lens correction is an unavoidable by product of all stitching programs. It's more a challenge to actually undo it's findings. As far as having different distortion characteristics at different magnification levels goes, I'm not sure but I don't think it's at all important to even try and create a profile with calibration software. My approach is to take an image with lots of detail and stitch it. Once the algorithm has aligned the mosaic I then check the results for the generated control points. I delete all of them with errors greater than 1 pixel. We know any such misalignment must be a fluke when stitching scans, or the film is buckled or something, which points to a problem other than parallax errors. Once the control points have been culled, re-optimize the image, don't create any new control points as you'll just be going backwards. If the preview looks good with the culled control points, it is good. At that point the lens will have been assigned a set of a, b, c distortion characteristics which will be all you'll need as far as calibration for future use. If you want to be really picky, push it further. Remove all control points greater than .7 and re-optimize. Then use the new lens distortion figures. These figures, at least in PTGui, can be saved as a preset. You now have a calbration for that magnification to assign (in PTGui it's 'load from database') future stitches.
Hats off to Helmut Dersch, the father of PanoTools, the progenitor of stitching programs.
Last edited by Daniel Moore; 23-Feb-2012 at 10:48. Reason: let's get some links in there