A stitching event

In another thread, http://www.openphotographyforums.com/forums/showthread.php?t=2865 there was discussion comparing the benefits of large format film compared to stitching of smaller format digital images. It was mentioned that the modern software allows easy and quick stitching of images to produce a digital image that can easily exceed the resolution of a lf film image. Bart gave a link to some software, which is indeed easy to use and can give accurate results if the initial images are carefully taken.

However, I believe the software is good enough to produce acceptable results with images that were not carefully taken, so I have started this thread hoping that we can learn just how this process can be so easy.

So, if you go open the directory at http://yertiz.com/images/pantest you will find 7 tif files, each file being about 1.2 MB in size. These are photos, taken earlier this year of an area of a formal garden. (They are tif images, your browser may not render them. You will need to download them to your pc in order to work on them.) The last image - all.tif - is composed of the other six images laid out in order, but obviously not stitched, and is provided to show the image sequence, and to indicate that the contrast is not consistent over the range (a common problem to be solved). They are six images taken from one position, iirc, with no particular intention to make a panorama. No processing has been done to the files, other than using Irfan view to convert the original cr2 file to tif, and down size them. The sky is nicely blown out...

The task is to use this software http://www.egelberg.se/panowizard/download.htm, or other low cost software to produce a panorama, and describe the steps involved to produce a stitched panorama. (I think panowizard may be pc based only, but I expect similar software exists for other platforms). There is no need to upload a high resolution result, since the idea is that we can follow your steps and produce a similar result on our own local machines, if we wish - it should be that simple. Even I was able to virtually immediately produce a good enough result to persuade me it was worth the effort to resize the images, etc., and upload them for you all to try.

I am not concerned with sharpening, or other manipulation, but I would like the resulting stitched image to show the path with the cypress trees to be straight, and not curved, if possible.

I'm hoping this will turn into a sort of tutorial, on various methods of stitching images, so if you just post your results, without saying the steps you took in order to get there (so that we can repeat the process), then you will have missed the real objective of this thread.

Best wishes,

Ray

PS - I think it was a guy named 'Chubby Checker' who had a hit single back in the 60's 'Let's stitch again, like we did last summer' (well if you can remember the 60's, you weren't there ;-))

Ray,

could you provide the used lens?
APP says, that the Pano's FOV is 165 degrees, to wide for a planar projection.

High Ray!

my take with Autopano (Mac beta version) Thanks to Michael Fontana, Klaus Esser, Bart_van_der_Wolf and others from the MF/stitch threads:



PSD available here

1 more row on top and 1 on bottom would have helped… to make a planar projection, this one is cylindar.

You have a very nice garden by the way ;-)

Hi Michael,

Unfortunately, Irfan view did not carry over the exif information from raw to tif format, however, below is the list of most of the Irfan view interpretation of the original raw file information.


_MG_4038.CR2

ImageWidth - 1536
ImageLength - 1024
BitsPerSample - 8 8 8
Compression - 6 (JPG)
Make - Canon
Model - Canon EOS 20D
StripOffset - 9728
Orientation - Top left
StripByteCount - 538470
XResolution - 72
YResolution - 72
ResolutionUnit - Inch
DateTime - 2007:02:10 13:54:25
ExifOffset - 270
ExposureTime - 1/200 seconds
FNumber - 6.30
ExposureProgram - Aperture priority
ISOSpeedRatings - 800
ExifVersion - 0221
DateTimeOriginal - 2007:02:10 13:54:25
DateTimeDigitized - 2007:02:10 13:54:25
ComponentsConfiguration -
ShutterSpeedValue - 1/200 seconds
ApertureValue - F 6.30
ExposureBiasValue - 0.67
MeteringMode - Partial
Flash - Not fired, compulsory flash mode
FocalLength - 17 mm
UserComment -
FlashPixVersion - 0100
ColorSpace - Uncalibrated (-1)
ExifImageWidth - 3504
ExifImageHeight - 2336
InteroperabilityOffset - 9332
FocalPlaneXResolution - 3959.32
FocalPlaneYResolution - 3959.32
FocalPlaneResolutionUnit - Inch
CustomRendered - Normal process
ExposureMode - Auto
White Balance - Auto
SceneCaptureType - Standard

Maker Note (Vendor): -
Macro mode - Normal
Self timer - Off
Quality - RAW
Flash mode - Not fired
Sequence mode - Single or Timer
Focus mode - One-Shot
Image size - Large
Easy shooting mode - Manual
Digital zoom - None
Contrast - High , +1
Saturation - High , +1
Sharpness - High , +1
ISO Value - 32767
Metering mode - Partial
Focus type - Auto
AF point selected -
Exposure mode - Av-priority
Focal length - 17 - 40 mm (1 mm)
Flash activity - Not fired
Flash details -
Focus mode 2 - Single
White Balance - Auto
Sequence number - 0
Flash bias - 1.00 EV
Subject Distance - 0
Image Type - Canon EOS 20D
Firmware Version - Firmware 2.0.3

I think the centre photo is darker, only due to the speed being at 1/640 second. I believe all the images were at 17mm focal length (20d is not full frame, but I think that makes no difference for this purpose.)

How many iof these images could APP accept to give a plano projection, will it allow, say, 3.5 images?

Best wishes,

Ray

Nicolas: Good one!

Ray: I gave a little stitch tool, like DoubleTake (mac only, 10 $) a try, as well as APP.

Click & save-tools can't handle that; more sophisticated ones, like APP make it.
One of the problems when stitching are moving trees, by wind. And with pict 4036, you were pretty out of the NNP.

So here's APP.






And DoubleTake:

No big difference by adding the lens & cam's (cropfactor) info.
This would be different with a FF-cam, as with a 1.6 crop, one uses the sweet spot of the lens, only.


Adding lens-info in APP:

Here are 3 quick tries with Photomerge in PSCS3. Had I taken these, they would have been in RAW and I would have corrected and synchroed before conversion--thinking it would have been even better.

The first is a 'reposition only' and cropped version (where it positions and blends the layers using masks).


The second one is the same before cropping

and the last is done requesting a cylindrical positioning and then blending


I've found this to work out quite well--and if you look at a thread on 'best landscape lens' you will find several stitched photos by Jack Flesher using Photomerge. I've used others but find this, esp. after conversion with RAW, to be as good as any others I've tried.

No tutorial required. I opened the tiff tiles in PSC3, went to file/automate/photomerge and clicked on 'add open files', selected my choice of ways to merge, and clicked on okay and let it go. I flattened, converted to sRGB and saved as jpeg.

Had I shot these, as I said, I would have shot in RAW (I realize these weren't shot for pano), then would have opened all in ACR4.1 together, corrected one of them to begin with and syncroed. I would then have checked to see if I felt that worked--if not, corrected each one, selected all and opened them in PSCS3 and Photomerged them. The Photomerge in 3 is so much better than in 2--worlds of difference to me.

Diane

Michael Fontana said:

Nicolas: Good one!

Click to expand...

Thanks Michael, but as you know, autopano makes most of it alone!

Some fast modifications can be done (wich I haven't) on the PSD file (all images stitched but not flattened).

It is a really amazing "little" software… and "smart blend" a good tool!

Nicolas Claris said:

Thanks Michael, but as you know, autopano makes most of it alone!

Some fast modifications can be done (wich I haven't) on the PSD file (all images stitched but not flattened).

It is a really amazing "little" software… and "smart blend" a good tool!

Click to expand...

yep Nicolas, "smart blend" was used for the first example, avoiding "overlapping" lines.

Diana, and Ray

PS- CS-3 photomerge vs these stich apps are different tools for different purposes.

For stitching once a year a landscape together, no special stich-apps are required.
This would be a overkill, in terms of learning curve, too.

If someone wants to use stitching regularly, asking for high definition, precise stitches, of buildings, interiors, etc stitching apps shine.

Forgotten the site with a comparison between these two possibilities.

Michael Fontana said:

Diana, and Ray

PS- CS-3 photomerge vs these stich apps are different tools for different purposes.

For stitching once a year a landscape together, no special stich-apps are required.
This would be a overkill, in terms of learning curve, too.

If someone wants to use stitching regularly, asking for high definition, precise stitches, of buildings, interiors, etc stitching apps shine.

Forgotten the site with a comparison between these two possibilities.

Click to expand...

I don't know if I'd agree with that generalization Michael...

The first Autopano example posted above has a nasty visible seam near the right hand frame bottom. By contrast, Dianne's last blend looks better, though it is smaller.

To be sure, Autopano is a great tool when captures are obtained carefully, but CS3's included compnent of automerge is pretty darn good for no cost...


Cheers,

Jack_Flesher said:

I don't know if I'd agree with that generalization Michael...

The first Autopano example posted above has a nasty visible seam near the right hand frame bottom. By contrast, Dianne's last blend looks better, though it is smaller.

To be sure, Autopano is a great tool when captures are obtained carefully, but CS3's included compnent of automerge is pretty darn good for no cost...


Cheers,

Click to expand...

Jack,

Nicolas didn't used the smart blend mode, therefore the seam.

> but CS3's included compnent of automerge is pretty darn good for no cost... <

I don't question this, and its progress since PS-CS-2, either.

My point: if you get in more complex situations, a plugin, even if it's working fine, can't have the features of a standalone, and - just one example - its capabilities of controlling, if necessary, manually the "dedecting points".

Some two weeks ago, I had to stich some shots taken with lots of wind. Clouds and trees moved a lot; therefore its "dedecting points" were bad. Any automatically generated stitch produces a (wrong, false) distortion of the image and its correct spots, to get the moving clouds and trees overlayered! The auto-apps have to interpolate between these errors, to avoid seams, and make it nice-looking.

But nice-looking might not be the goal for these stitches, I mentioned before:

Only removing the wrong "dedecting points" - which can be done in APP with a mouseclick, will create the geometrical correct pano. Off course, it takes longer.

I hope my english is gooed enough to make that clear

Michael Fontana said:

Jack,

Nicolas didn't used the smart blend mode, therefore the seam.

> but CS3's included compnent of automerge is pretty darn good for no cost... <

I don't question this, and its progress since PS-CS-2, either.

My point: if you get in more complex situations, a plugin, even if it's working fine, can't have the features of a standalone, and - just one example - its capabilities of controlling, if necessary, manually the "dedecting points".

Some two weeks ago, I had to stich some shots taken with lots of wind. Clouds and trees moved a lot; therefore its "dedecting points" were bad. Any automatically generated stitch produces a (wrong, false) distortion of the image and its correct spots, to get the moving clouds and trees overlayered! The auto-apps have to interpolate between these errors, to avoid seams, and make it nice-looking.

But nice-looking might not be the goal for these stitches, I mentioned before:

Only removing the wrong "dedecting points" - which can be done in APP with a mouseclick, will create the geometrical correct pano. Off course, it takes longer.

I hope my english is gooed enough to make that clear

Click to expand...

Yes indeed, I understand your point now and agree

Cheers,

Michael,

Not all the APP interface is so intuitive. I must admit was a little flummexed by the presentation all the focus points! Where did you find the methodology?

Asher

Michael
yes I did use the smartblend!
here after 3 minutes only editing the PSD file that I've already posted upthere, + a little sharpening (sorry, I can't leave soft image…)



One can nearly breath the fresh and wet British lawn grass perfume!

Michael Fontana said:

My point: if you get in more complex situations, a plugin, even if it's working fine, can't have the features of a standalone, and - just one example - its capabilities of controlling, if necessary, manually the "dedecting points".

Click to expand...

And there are other differences as well, like handling exceptional cases. Sofar, none of the posters addressed Ray's
desire "... but I would like the resulting stitched image to show the path with the cypress trees to be straight, and not
curved, if possible".

Now, while that is not too difficult with a tool that allows to tweak lots of details, it does come at a price; loss of
resolution. However, if the original has enough resolution there is a lot you can get away with.

So here's my attempt (it's not perfect, just a quicky) with Ray's specific request in mind:



And this is the full uncropped stitch it comes from:


As you can see, a significant projection distortion was needed to (more or less) straighten the path. I used
PTAssembler's "Straight-line" tool, where you draw along a straight feature and the software does the rest while respecting the other image overlaps.

Bart

Bart,

This, to me is much more satisfying. I like that you have put the world in contest. Well down!
For Mac users, how do you do this in Photoshop?

Nicolas,

Tell me what you did in your last presented image: converted to sRGB or assigned as the colors are again over-saturated, to my judgement, at leadst!

Asher

oummm, Nicolas, here we go

that line at the right, in your initailally posted stitch looks to me like to be a classical blending "error", as i t didn't apperead a my example.

For some infos, french only, look here
Now, I found a english one: http://en.wiki.autopano.net/Interpolation_and_blenders

Asher, what do you intent by "focus points" ??

You might find the control points in the Control point's editor.

http://en.wiki.autopano.net/Control_Points_Editor

Michael Fontana said:

that line at the right, in your initailally posted stitch looks to me like to be a classical blending "error", as i t didn't apperead a my example.

Click to expand...

I agree with you, but I got rid of it very easily in the 2nd attempt.
However this is not a nice smart blend behavior, I guess it is maybe because of the Beta version I use on Mac.
We also have now APP running on the PC (for the 360°) and dnoticed that the renderings are not the same between Mac and PV version… Beta oh! Beta…

Asher, no I did not do anything to color space exept export for the web that includes sRGB.
I thing the color saturation may increase from the auto-color (wich I did enable) of APP and also the midtone sharpening from Light and Shadow filter which I may be too intensly used to enhance light zones, but this OT…

Bart_van_der_Wolf said:

And there are other differences as well, like handling exceptional cases. Sofar, none of the posters addressed Ray's
desire "... but I would like the resulting stitched image to show the path with the cypress trees to be straight, and not
curved, if possible".

Click to expand...

Bart

this is exactly why I wrote in my 1st post:

1 more row on top and 1 on bottom would have helped… to make a planar projection, this one is cylindar.

Click to expand...

And Michael as well:

APP says, that the Pano's FOV is 165 degrees, to wide for a planar projection.

Click to expand...

IMO, planar projection works well till 100 degrees, maybe 120 depending on the subject.

At 180 deg, the lines at the left and right borders go towards infinity; therefore at 160 already, these nasty distortions already appear. Physics, a lens can't go wider than 180 ;-)


There's one workarround until jet, only, the multinode-pano, pretty at the edge of today's science:
Far to complex for that thread, but if interested, you might look at that video

Nocolas: it's not even a beta, its a alpha

Asher Kelman said:

Bart,

This, to me is much more satisfying.

Click to expand...

Yes, it looks more natural, but Ray's samples covered such a large FOV that normal rectangular projection can't really deal with it. Normal rectangular projections work up to about 120 degrees, beyond that one needs to open the bag of tricks/tools.

For Mac users, how do you do this in Photoshop?

Click to expand...

Photoshop (CS3 !) does a reasonably good job with regular stitches, and the blending is not bad although it can be improved (there are also undocumented? blending tricks). I have yet to find the solution for the straight features in Photoshop CS3 (installed it last night, so I'm not yet fully familiar with all the latest features).

Windows, Mac, and Linux users, can use Hugin if they want to have more control, such as with the "Straight line" feature. There is a new stable version about to be released (presumably also as precompiled binaries).

Bart

Michael Fontana said:

For some infos, french only, look here
Now, I found a english one: http://en.wiki.autopano.net/Interpolation_and_blenders
You might find the control points in the Control point's editor.

http://en.wiki.autopano.net/Control_Points_Editor

Click to expand...

Michael
you already gave me these links before, but I hadn't time to read… so stupid, what a time saver!

It's all in French there

Thank you beaucoup !

I uploaded some shots that will be rather be difficult to stitch; the lens is a sigma 24 mm macro, with lots of distortion, but it's the only widemacro, I know. I made the test to see, if I could use that old lens for close macrostitches; as the 24 mm should provide a better DOF, than longer macro's.

Difficult to stich too, as the object is pretty close to the lens: 18 cm!






We might use these 3 jpgs for a stitching apps comparison, they ' re reduced a bit in size, but straight out of the converter, download them here

PTAssembler attempt

Michael Fontana said:

Difficult to stich too, as the object is pretty close to the lens: 18 cm!

Click to expand...

Yes, not as simple as it seemed, but do-able:


It could do with some further (overall) distortion correction and further tweaking, but that's always required when the initial camera rotation setup is not optimal. It can also help to separately 'calibrate' the lens with the Pano software for its specific distortions beforehand.

Bart

well done, Bart!

> calibrate the lens prior to stitch..<
that's exactly what I thought also, as I had little misalignements as well.

Thanks for having a go. Some interesting results. I had hoped an explanation of how it was done, with the low cost solutions would encourage others to try with photos they may have - basically any photos with overlapping views. I think Diane was the only one to give the steps involved.

The PTassembler solution, as shown by Bart, in this instance gives a reasonable result. Even working with the jpeg as shown, printed at about 48 inches by 8 inches, provided you don't look too closely, is 'nice'. Cropping both sides, to half way through the end cypress trees, improves the composition, a small amount of post processing to remove the distorted circular flower borders at the ends, add some sky, would possibly be worth while, if working with the raw image. However, how to get there, what are the steps in using the software, is it simple, or time consuming? Can APP get it straight?

Perhaps I will try with Michael's photos, but they were taken with the view of trying to stitch, so they should give a more accurate result.

Best wishes,

Ray

Good day, Ray

>However, how to get there, what are the steps in using the software, is it simple, or time consuming?.....<

More the situation is complex, and a higher precision is required, more one needs the appropriate tool, and they include a lerning curve: it's like just picture-editing.

>Can APP get it straight?<
Nicolas's and my example of the cemetery were straight; the box, not beeing in the NoNodalPoint, can't be done without the big tools: as closer you go, the problems will raise.

Basically, it's up to you, to decide how precise and how often you want to stich difficult situations.
Being kinda MF-back-replacement to me, I tried to learn it from the bottom; as shooting lots of architecture; I am obviously looking for the highest precision I can get. It took me about 20 stitches, to understand APP fully; it it's not a click and forget-app in difficult situations, but it has some automated routines, as autodedection of control points, for the easy stitches. BTW: Other stitching apps might have this as well.


Bart, when calibrating the lens prior to stitch, the results were worse!

Ray West said:

Thanks for having a go. Some interesting results. I had hoped an explanation of how it was done, with the low cost solutions would encourage others to try with photos they may have - basically any photos with overlapping views. I think Diane was the only one to give the steps involved.

Click to expand...

Hi Ray,

I started to document my steps with PTAssembler, but it takes longer to decribe than to do. Also, while it offers all sorts of automation, PTAssembler is a highly tweakable tool geared at large gigapixel projects (but also other sorts of stitching). The possibilities are endless, but the learning curve is therefore a bit steeper than e.g. AutoPano Pro. So describing the setup parameters, including the installation options, would not be very inviting to casual stitchers but rather scare them away. I only wanted to demonstrate something different, including the result, from what others have produced with a more automated approach.

The PTassembler solution, as shown by Bart, in this instance gives a reasonable result. Even working with the jpeg as shown, printed at about 48 inches by 8 inches, provided you don't look too closely, is 'nice'. Cropping both sides, to half way through the end cypress trees, improves the composition, a small amount of post processing to remove the distorted circular flower borders at the ends, add some sky, would possibly be worth while, if working with the raw image.

Click to expand...

The output could have been better if the input image quality had been adjusted before stitching, but that as not the challenge. One of the goals was to see how the various stitchers blended the exposure differences, and the imperfect shooting conditions. Also, the extreme FOV (and the desire for straight lines) caused image degradation. My advice to people is to always shoot subjects that don't fit in a single shot, as a potential stitched panorama. Afterall, there are possibilities to recreate the full scene, and with good input image quality, the end result can be exceptionally good (FOV and/or resolution wise). Stitching opens an array of possibilities to get shots that are otherwise impossible, or at least very difficult/expensive, to achieve with a single shot.

However, how to get there, what are the steps in using the software, is it simple, or time consuming? Can APP get it straight?

Click to expand...

PTAssembler will get the job done, but may require human intervention for pathological cases. That intervention requires skill/experience from the operator. Skill takes time to develop. The basic operation of PTA is simple enough; import the images, click on auto and view the result. However, the result will often not be good when the input images are not good (not made with the intenton to stitch).

Whether APP can get it straight is sofar an open question, maybe I should give it a whirl.

Perhaps I will try with Michael's photos, but they were taken with the view of trying to stitch, so they should give a more accurate result.

Click to expand...

From a blending perspective it was a much easier job, but there are some pitfalls in getting it to stitch without visible errors. The lens distortion correction is solvable, but there are setup issues (not rotated through the entry pupil of the lens, causing close-up parallax errors) that cause suboptimal output (too many impossible compromises).

Again, my advice for all is to learn the important factors for successfull stitching (by playing with it) so one is prepared to shoot with a good enough quality (even handheld) to save an otherwise lost opportunity.

It is also clever to shoot with a longer focal length in order to gain resolution, even if that means that the subject doesn't fit in one frame. That is e.g. why I shot the Berlin Cathedral (here at 33% of final 12 MP output dimensions) zoomed in, instead of wide-angle, on my handheld Powershot G3 requiring 6 shots to cover the FOV. Turning a 4 MP point-and-shoot into an up to 62 MP wide angle (28mm in 35mm equivalent FOV) camera is not bad in my view ... , it just requires some postprocessing.

Bart

Michael Fontana said:

Bart, when calibrating the lens prior to stitch, the results were worse!

Click to expand...

Yes, I guess the parallax error is virtually impossible to remove, due to close-up shooting without an optimized rotation platform. But if it would be a really important image one could spend some more time getting it closer to perfect. PTAssembler also allows to morph control points into compliance.

Bart