Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: sharperstill on July 19, 2013, 02:08:00 am
-
Hi,
I've not done a huge amount of stitched panoramas in my life, but have found them to be not too problematic until now.
I was road testing my new Canon TS 17mm and thought I'd womp together a stitched pano at a nearby swamp to see how it'd look.
See attached, and notice the two distinctive bumps on the horizon line. The power lines to the right of the nearest tower are also very mis-aligned.
I was reasonably careful when levelling the tripod head.
Any clues where I let this one get away?
The two version are two different panorama modes in Photoshop, but virtually identical.
Jon
-
Hi,
I've not done a huge amount of stitched panoramas in my life, but have found them to be not too problematic until now.
I was road testing my new Canon TS 17mm and thought I'd womp together a stitched pano at a nearby swamp to see how it'd look.
See attached, and notice the two distinctive bumps on the horizon line. The power lines to the right of the nearest tower are also very mis-aligned.
I was reasonably careful when levelling the tripod head.
Any clues where I let this one get away?
Hi Jon,
Hard to say, because of the black-box approach of Photoshop. In my experience, which is extensive when it comes to Pano-stitching, Photoshop can do an okay kind of job, as long as the input is not too problematic/challenging.
Input can be problematic if there is no EXIF data, or the images have been cropped or distorted before stitching. Photoshop can also have a problem with Tilt and Shift lenses, especially with when shift was applied. Also when there is very little overlap between the images and they were not rotated through the axes at the location of the entrance pupil of the lens, it may be impossible to align both foreground and background detail.
A dedicated stitching program should not have too much problem with most of that, and even leveling should not be absolutely necessary, also because the program would allow to manually tweak the parameters that need to be used to get a better result. Photoshop doesn't allow adjustment of parameters, so if it works it works, but if it doesn't, you're usually stuck. All that then remains is editing with e.g. the Warp tools.
If the problem you experience is caused by missing EXIF data, you could try and paste a copy of the image in an image that has the same camera settings, especially knowing the correct focal length is important for stitching. If the problem is caused by (some of) the images having shift applied, you can pre-process the image tiles by adding some transparent Canvas space to each one of them, enough to vertically center the real horizon in each image.
For problem images, you could also try a free very capable dedicated stitching program such as Hugin, which does allow lots of manual intervention. It can also use better resampling algorithms than Photoshop, and that will produce sharper output.
Cheers,
Bart
-
could be distortion from the lens - see if the lens distortion correction makes a difference.
could be mismatched stitching - maybe ptgui would do better as you can define the stitching points.
keith
-
Perhaps a silly question but are the 'bumps' actually represented in the geography?
Another question: Are the seams of the stitches overlapping the 'bumps'?
You did not mention in the original post how many images were used to stitch the panorama.
If there were only two then those bumps are not due to the stitching.
Three images stitched may be responsible.
From a technical perspective you also did not mention whether you rotated the lens or used the shift ability of this lens.
Also, you may have leveled the lens and camera appropriately but that in itself does not guarantee a good result.
I think that if you explain your technique in enough detail we could probably work it out but currently there are so many variables involving both shooting and post-production as to make it impossible to guess.
Tony Jay
-
You should try autopano pro of PT gui.
Cheers,
Bernard
-
A typical and quite annoying quirk of PS' Photomerge, as far as I can see.
Go to a better panorama stitcher - one could add Hugin to Bernard's answer.
-
You might want to try 'Puppet Warp' to the stitched image to correct the humps.
-
In french, we'd call that a plaster on a pegleg. ;)
But yes, I'm afraid it could be the only hope within PS.
-
Sometimes you can get 'interesting' results!
http://www.wolfnowl.com/2010/05/panoramic-photography-and-stitching-errors/
Mike.
-
Sometimes you can get 'interesting' results!
http://www.wolfnowl.com/2010/05/panoramic-photography-and-stitching-errors/
Mike.
Gosh, Mike! You've got a great new creative tool there! The "fix" on the second shot is my favorite.
Eric M.
-
It would be useful to know how many frames you used, and what was the overlap.
As mentioned by another poster, it could be caused by the lens distortion. If you can repeat the stitching experiment, you might find your answer by looking at the individual layers and then correlate the stitching lines with the problem areas in the final image.
Most stitching programs put priority on the alignment of individual segments, not taking into consideration the lens distortion or blurry edges.
For example, they may use the leftmost and rightmost images almost in their entire lengths, while discarding a lot of information from the neighbouring frames with much sharper and less distorted data.
To counteract it, sometimes I crop the individual images on both sides, thus eliminating the greatest distortions, and forcing the stitching program to use only the best portion of each frame.
-
Thanks for the replies everyone and sorry for the delay in responding - blame the time zone and a busier than expected day yesterday.
Some points raised by respondents.
It was 4 images used to generate the panorama and overlapped a third of the frame
The lens was shifted upward slightly. I can't remember the exact amount but at a guess probably shifted half-way. There was no tilt applied.
The camera was positioned on a 'nodal slide' in a position I had worked out earlier from the two-candles-on-my-dining-table method.
"Input can be problematic if there is no EXIF data" - The file is on my other computer, but I imagine the EXIF data was intact. The RAW files were imported into Lightroom then all four selected and sent to Photoshop to Merge to Panorama.
"could be distortion from the lens - see if the lens distortion correction makes a difference." - I didn't think lens corrections could be applied to images from TS lenses?
"Perhaps a silly question but are the 'bumps' actually represented in the geography?" - No, it's pretty flat, it's a wetland...
I ditched the panoramas after outputting the lo-res composite but had a look at the masks that Photoshop had made before doing so. They were weird. Previous panoramas (3 frames) that I've done with the 24TS PS used each image in one 'slice' but in this one there was one frame used where PS had selected several separate 'islands' of data from that frame only, including two that were really quite small and of insignificant areas.
I've downloaded Hugin, and will process the files in that and report back.
Cheers,
Jon
-
The lens was shifted upward slightly.
Hi Jon,
Well, as expected there's your problem, it confuses Photoshop.
I've downloaded Hugin, and will process the files in that and report back.
Once you've gotten a bit acquainted with the application, the solution for your issue is in the 'y shift (e):' parameter of the Optimizer Tab. Typically, you'd do a regular stitching optimization (which may get similarly confused as Photoshop), switch-off the optimization for all other parameters, then only optimize the 'y shift (e)' parameter, switch it off again, and do another regular optimization of only the other parameters again.
Once things improve, you can be so bold as to optimize the lot of parameters, but with so many variables, it can also result in total chaos. So make sure you save the project file between optimizations, before you accidentally lose the Undo capabilities ... It's not a question of if, but when ...
When you make a note of the shift quantity applied to the lens, it's easy to transfer that amount to the optimization parameter. A shift of 1mm on a 36mm sensor in portrait orientation, leads to an image pixels in height x 1/36 offset. You'll have to figure out if it's a + or - offset depending on the shift direction used.
Also note that, for a dedicated stitching application it is not absolutely necessary to shoot level and compensate for key-stoning or such by shifting the lens. It can help to get a better coverage of the sensor real-estate because no cropping is required afterwards, but mathematically it make no difference for the Pano-Stitcher.
Cheers,
Bart
-
Thanks Bart,
Definitely sounds less 'black box' than Photoshop. I'll give the same files a go over the next day or two.
Jon
-
Thanks Bart,
Definitely sounds less 'black box' than Photoshop. I'll give the same files a go over the next day or two.
Hi Jon,
Absolutely less 'black box'. But the downside to more control is also more opportunities to make an even bigger mess. Hence the practical hints.
Cheers,
Bart
-
Quite astonishing how the canard of "tripod levelling" recurs when stitched panos are discussed. As if, within reason, it mattered when you are remapping every pixel in the final image. As far as I'm aware all stitching software (I'm only familiar with PTGui Pro) has the inherent capacity to level the final image.
Roy
-
Photomerge has gotten better with each iteration of Photoshop and it is to the point now where I try Photomerge in Photoshop CC before turning to PTGui Pro. Photomerge is still mostly a simpler "blackbox" as far as fine tuning the controls while PTGui Pro approaches infinity in terms of ability to tweak the workings.
If you have PsCS6 or Ps CC check out this video by Russell Brown on the Adaptive Wide Angle filter: http://www.youtube.com/watch?v=0LUlpj4DqIo at 7;12 he directly discusses using it with panoramas and why he uses it now instead of Puppet Warp
-
I have done extensive stitching using my cell phone camera for various site related jobs. I have used both Hugin and Microsoft ICE. The problem with bumps always occurs on both when
. Taking a pano of a long beam head on - distance is quite short
. Taking a pano of a long wall head on
I have tried 20% to 80% overlap, but nothing helps in either of these software. The only remedy is to tape panorama shots at an angle, or use a software where you can specify a horizontal/vertical alignment line
You can see that there is a disjoint in the centre
-
I have tried 20% to 80% overlap, but nothing helps in either of these software. The only remedy is to tape panorama shots at an angle, or use a software where you can specify a horizontal/vertical alignment line
Hi,
Hugin allows to specify horizontal/vertical lines. When the images were taken with some care, even handheld images can often be stitched by using the offset parameters (Image center shift: d and e), and for flat surfaces the X, Y, Z parameters.
Cheers,
Bart
-
Ajoy Roy,
I have tried...but nothing helps...
Did you place control points manually?
-
Have you tried the Photoshop's Photomerge "Reposition" (last option in the list)? It almost always works best for my wide angle TS-E shots when I use an Induro PHQ-3 head.
Ken
-
I'm back.
I'm happy to report I got a good-looking-no-bumps panorama out of Hugin. See attached.
The only bad news is I'm not really sure how I did it.
My forst attempt with Hugin yielded much the same result as in the original post. I left it for while, got busy with other things and came back to it yesterday.
Starting from scratch and using only 3 images rather than 4, for roughly the same field of view, I got the result shown.
Obviously I'm still groping my way around Hugin like a sailor in a brothel but I'm pleased with the outcome.
Jon
-
I'm back.
I'm happy to report I got a good-looking-no-bumps panorama out of Hugin. See attached.
The only bad news is I'm not really sure how I did it.
Hi Jon,
Good to know it worked out as intended.
From the looks of it, you've selected a cylindrical projection, which makes sense with such a wide field of view. That does mean that only a centrally placed horizon will be straight, and that's what you seem to have done, vertically centered on the horizon.
You can also achieve that automatically, even when your image tiles have more sky (or more land) than land (or more sky), by placing some horizontal control point pairs in each image tile, if it's easy to see that horizon like in your images. By adding those horizontal control point pairs in each individual image tile, you'll automatically force the image to a higher or lower than center image, and subsequent cropping will take care of the rest. It may cause empty areas in the outermost corners, and/or the center, so one might be required to take some additional images for those areas if the scene is predominantly sky or land.
Other projection methods (http://www.tawbaware.com/projections.htm) offer other opportunities, but also challenges. Not all projections are available in every Pano-stitching application, some only offer a few.
Cheers,
Bart
-
Sometimes you can get 'interesting' results!
http://www.wolfnowl.com/2010/05/panoramic-photography-and-stitching-errors/
Mike.
Similar things have happened to my panoramas when taken with extreme wide-angles. In the past there was a recommendation that the lens used for stitch panoramas should not be wider than something like 28mm for 135 size sensors. There is so much distortion in wide-angle corners that the stitching software is bound to have problems sometimes. They have been getting better in this respect, though. 17mm is wide...
-
" ...I didn't think lens corrections could be applied to images from TS lenses?..."
Hai Jon, to my knowledge , the only program that can do lens corrections on TS lenses is PTlens
http://epaperpress.com/ptlens/
good luck!
Pieter Kers
-
Petrus,
In the past there was a recommendation that the lens used for stitch panoramas should not be wider than something like 28mm for 135 size sensors. There is so much distortion in wide-angle corners that the stitching software is bound to have problems sometimes.
If control points are placed manually there is no reason to expect anything less than a perfect stitch from an ultrawide angle lens. Distortion is taken into account by stitching software for any lens - it has to in order to theoretically overlap each image exactly with its neighbour(s).
In my opinion most stitching errors derive from the user allowing the software to place control points. Wide angle lens or otherwise.
-
Petrus,
If control points are placed manually there is no reason to expect anything less than a perfect stitch from an ultrawide angle lens. Distortion is taken into account by stitching software for any lens - it has to in order to theoretically overlap each image exactly with its neighbour(s).
Hi,
That's correct.
In my opinion most stitching errors derive from the user allowing the software to place control points. Wide angle lens or otherwise.
While true, and that's why I always double check the individual control points when there is an issue getting the stitch to work well, there are other reasons as well. The optimization process can get stuck in a mathematically optimal solution that minimizes the errors, but given the complexity of the problem there may be other optimal solutions as well, and some give better results.
It can therefore make a difference in what order the various parameters are optimized, and some programs allow to influence that. Different orders can be required for stitching rectilinear panos or spherical VR panos, or for wider angle lenses or for normal to tele-lenses.
Cheers,
Bart
-
Auto pano pro does a much better job of stitching fine detail than photoshop.