Phun with Photomerge

This Photomerge capability in Photoshop is really quite fascinating. There's a tree near my house that keeps striking me as pretty with the afternoon sun on it, but I've never been able to photograph it because I don't have nearly a wide enough lens—and if I step back farther, the view of it is blocked by nearer trees. So just as a lark I went out yesterday and shot off nine frames with the Pentax K20D and 35mm ƒ/2.8 Macro. I made the shots handheld from quite near the tree—I stood maybe twenty feet from the stop sign on the right (see the last illustration below)—and I didn't do a very good job at all, it turns out, of keeping my spacing consistent. Then I selected all nine pictures and let Photomerge chew on the data for a while.

For those of you who've never tried it and are wondering how to do it, it's very easy in Elements 6: you just go to file —> new —> Photomerge panorama, then choose the pictures you want to put together, and it does its thing automatically. It takes a while, but it's not hard.

First of all, here's a small map of the individual frames and how Photomerge "jigsawed" them together. I can't make this bigger, but maybe it's big enough for you to see (the second illustration from the bottom of this post is the third one down here):

There were some anomalies. All I did to the merged image you see below was to crop it and straighten it slightly, but after the crop, a few mis-fits in the puzzle pieces showed up. Here's an example from the lower left-hand side of the image:

Oddly, these don't show up in the 100% version, but become quite prominent when I reduce the image size. I wonder if that's because I didn't flatten the layers in the .psd file first? (I'm not quite ready to do that yet.)

There are a couple of mis-stitches, too, for instance here:

...and some of them are relatively important, too (look at the tree and the telephone pole). I think I may have overlapped the frames too much. Here's a single shot to put you in context as to what my lens's angle of view is (it's about equivalent to a 50mm lens on a 35mm camera):

And finally, here's the raw result of the nine shots as put together by the Photomerge function. I think this is about equivalent to the angle of view of a 24mm-e lens, maybe as wide as a 21mm-e. (Click on the image to see it a little bigger.)

Mimicking a wider lens is probably not the software's strong point—it would probably be a lot better just to use a wider lens, which I didn't have on hand. Still, flaws and all, the software did a pretty creditable job, especially considering that I may not know how best to give it what it needs.

Photography really is as much about software as it is about hardware now. Photomerge at least goes a ways toward eliminating some of the limitations of shooting with a prime lens that's a little too long, which the DA 35mm Macro is for me. It makes it easy to change the aspect ratio of an image—you could fairly easily "shoot square" with a regular 3:2 ratio camera in many situations, for instance—and it gives you the option of a much wider angle of view than the native angle of the lens you're carrying, which could come in handy from time to time, as it did for me on Tuesday when trying to convey the scope of the flooded Fox River. It's not as good as a real panoramic camera because you have to have a certain stasis in your subject matter (you could have action in one or more frames, but not across the stitching boundary). Then again, with a little practice you wouldn't need a tripod, and some kinds of panoramic cameras—the scanning types like the Widelux—don't have instantaneous capture either.

I'm sure a lot of you reading this are a lot further up the learning curve with this than I am (this is all of my second attempt), just as, for some of you, it's all old news. And today's picture is a failed attempt, I think, all things considered (it doesn't really convey the looming quality of the tree at that corner, for one thing). But I'm impressed with the functionality. That I've never shot panoramas before has been mainly an issue of logistics: as a typical hobbyist I haven't wanted to deal with the expense of a separate panoramic camera or the burden of carrying it. But with neither of those penalties in play, that game has changed. I'm quite sure I'll be doing two- and three-frame panoramas in the future, at least occasionally, for subjects that warrant it. 

______________________

Mike

Featured Comment by Ctein: "You can deal with most of the stitching glitches you're seeing by choosing  'interactive layout' in the photomerge control panel. Once the software has loaded all the images, it brings up a working window with the images arranged in the way it thinks is correct (similar to auto mode). Then you can drag them around and reposition them where you want. Photomerge will still snap them into alignment when you've got it close to correct, but you get to decide the starting point that is 'close' instead of the software doing it (erroneously).

"Note that the stitching errors are not addressed by rotating the camera about the nodal point nor using a special panoramic head. They are software glitches in the pattern analysis. Out in the real world I have  not noticed any difference in quality from the precise manner in which the panoramic photos are made. The ones I showed of the Anchorage skyline in my 'stress tests' in previous columns were made hanging out a hotel window with the camera handheld; there wasn't anything close to trying maintaining proper body alignment.

"The analysis algorithms are extremely sophisticated and do sometimes make mistakes. For example, in the telephone line enlargement you showed, along with a mismatch there is what looks like a 'fold' in the  sky like a sheet of paper had been creased. I saw that happen in my super-large panoramic in a completely blank part of the sky midway up off to the left side. Photomerge had stitched the images together perfectly automatically, so this was very strange. It looked like there was a crease in my monitor! Easy enough to retouch out in Photoshop, but I have no idea what caused it!

"I have also on occasion had Photomerge change its mind. One of my stress tests rendered correctly automatically the first two times I ran it. The third time I ran it the software put the individual images together in the wrong order. And thereafter it always put them together in the wrong order; I couldn't figure out how to convince it to automatically put them together correctly again (interactive layout solved that).  Why it decided to do it differently on different occasions? I have no idea.

"This is why we really don't want artificial intelligence. A.I. gets notions!"

Featured Comment by Michael Bishop: "Here is my take on Photomerge; deconstruct it!"

Comments

[William Poole]

Undoubtedly many folks know more about optics than I do--and maybe they will comment--but "mimic a wider lens" does not seem exactly right to me. I regularly use photomerge to combine images made with a 45 mm TSE lens on a tripod. I make three "landscape" aspect images at the three shift points of the lens--fully shifted down, unshifted, and fully shifted up. This yields a single "portrait" aspect image of significant file size and nearly perfect registration between the individual files. What strikes me about the images is not just that they have a wider field, but that they also preserve the other optical characteristics of a near-normal lens--for example the relationship of image elements front to back--without the feel of a wide angle. I guess the same is true of stitched images made with any lens to cover the angle of view of a shorter one. I think this is why stiched panoramas have a different look than if the image was made with a wide angle and then cropped to the same area. Or maybe I have this all wrong.

Bill Poole

[Mike J.]

Bill,
No, I don't think you have it wrong at all. I just don't know what the differences would be, exactly, having never seen them side-by-side to compare. Photomerge with a TS lens--now that is an awesome prospect.

Mike J.

[Dierk]

The slivers you see in anything but 100% view are, in my experience, purely monitor rendering issues, no real misfits.

[Mario Alessi]

If it looks fine at 100%, it will look fine when flattened. When you zoom out photoshop must approximate what it will look like. Generally I find that quarter steps down look better on the screen than third steps for instance. 50% or 25% looks bette on screen than 33% or 66%. And I have found when you flatten it does look proper as long as it did at 100%. Also if you rotate around the "nodal point" of the lens you will have less distortion.

[Thiago Silva]

Hey Mike,

I happened to me the other day too, I merged a panorama and got those weird lines. But they are a display artifact only. If you flatten the image, they disappear.

Cheers!

[Rory]

Dave Beckerman (my other daily read) has been playing about with photomerge too - see http://beckermanphoto.com/2008/05/28/park-avenue-spring/ for example. I don't have photoshop, so can't really comment, but I'm interested to see that it can stitch verically as well as horizontally - up till now I thought software could only stitch photos from left to right or vice versa

[Nicolas]

As said, the "cracks" at odd magnifications are display artifacts and don't show up in the final flattened file.

Comparing with "stitched" images from a tilting camera, the main difference would be projection - it all depends on what you choose in Photomerge (I generally avoid the auto setting as it seemed less efficient on some cases), but in a nutshell rectilinear (the projection of the tilt&stitch solution) may not be that pretty if you got to go really wide...
For that matter, getting a feet into a real stitcher (Hugin comes to my mind, but it's not alone, and I've heard good things about PTGui too) may allow you to experiment a bit more.

[Nicolas]

Oh, and about the stitching glitches, there is one solution : it's called a pano head (eg panosaurus, nodal ninja...). Otherwise, mimick it while hand-holding the camera : try to move it around the entrance pupil of the lens. Easier to say than to do, with wide-angles.

[Kjetil]

Well, arent't you closer to mimicking the same focal length on a larger film format with the TSE? Anyway - for static subjects, the flexibility that can be added, especially to primes, by using this method, is very inspiring. I'm still like a little child on Christmas eve while watching the image come together in Autopano, the results provided by these algorithms fascinate me..

[Marco Quartulli]

I for one agree with Bill Poole.

Tele lenses have less scene angle coverage (acquire a thinner 'conical' portion of the whole sphere centered on them) than wide angle ones (larger scene coverage).
Images taken with long tele lenses will therefore 'look awkward' if the viewer is forced to observe them in an arrangement in which they cover a very very large portion of the visible sphere centered in the eye.

Therefore, for a viewer at a fixed distance from the print, tele images would typically be printed in smaller sizes than wide-angle ones in order to keep the geometries coherent - between acquisition (3d to 2d - by the camera) and vision/reconstruction (2d to a perceived 'normality' in the inferrable 3d).

For example: at a recent exibition at the Museo in Trastevere in Rome, a very large print of a telephoto had been (most probably on purpose) set up on a wall in a narrow corridor. Everyone I asked confirmed they could not get far enough from the print to 'see it right'.
Similarly, I suspect movies for fruition on modern large LCD and plasma screens are shot with wide lenses to give the not-too-distant spectator a comfortable sensation of full visual immersion.

This is of course not a digital thing in itself - it works perfectly well with juxtaposed 'traditional' photos.

In summary: focal length kind of acts as a measure of typical print size for comfortable hand-held observation.
Crucial experiment: a very long stiched strip of 50mm pictures will be instinctively be placed at exactly the same viewing distance as one of the pictures alone (I tried :-)).

Then: was fixed-lens 50mm 'normal' photography more balanced and classical-looking than zoom-based one for magazines and books constructed around full-page, fixed-area pictures?

[Rafael Galliez]

When I work with panomara I prefer do use the free software HUGIN, and compared with photoshop the result are much better. Maybe you could try with the same pictures and talk about your experience.

[Jack Nelson]

Mike,

I've done quite a bit of photomerge, both to make panoramas and also as a virtual 30+ megabit camera. I've found that before you straighten out or resize an image it's best to flatten it; I also keep an un-flattened copy. Before flattening it I turn the layers off and on and check for parallax errors. If the stitching is done from a distance these are minimal, but the closer you get to the subject the more likely it is that things will not line up. It is generally not too hard to correct these errors using either the lasso tool or the clone tool.

I do a lot of photomerge in a three by three grid; three across and three up; thereby increasing the pixel count substantially. I can then print large (12 x 24") prints from my 6.1 megapixel Nikon D50.

You can see examples here:

http://jack-nelson.com/photos/sagradoprofanado/pages/01.html

All ten of the images in this gallery were stitched, with the smallest ones being 9 exposures (3 x 3).

[Mike C.]

I'm surprised you haven't tried panoramic "stitching" before -- Panorama Factory and other software will do a bang-up job if you do the proper preparation. Above all, (a) use a tripod, (b) level your camera and (c) find the nodal point of your lens (probably not where the tripod socket is, and varying for zoom settings) and come up with a means of offsetting the camera so that, mounted on a tripod, it rotates around this point. That's about it... There are various ways of finding the nodal point, generally by lining up two vertical edges and rotating the camera: once they stay lined up in the viewfinder, you've found it. 3-D rotation is more complex, which is why you can buy expensive "panoramic heads" -- there's a whole web culture devoted to panoramic DIY self-help, though. I think I can guarantee you'll get bored with it pretty fast, though...

[Jeff Baker]

I've had trouble using PhotoShop's merge function with hand-held shots (i.e. anything not using a tripod, pan head, metered movements, etc). T/S lenses are a nice option, but you still need to bring a tripod, right?

I found this free software called autostitch (autostitch.net) that seems to work magic with almost every set of pan photos I've thrown at it. I've had PhotoShop work on things autostitch doesn't like, but overall, autostitch is my favorite.

Maybe it'll clean up the wire issue. It is annoying how supposedly smart software can't handle really obvious mismatches like that, but I guess it's only human.

[Dave Feltenberger]

Michael -

I wonder if Photoshop CS3 has a different engine than PS Elements? My first attempt at a panorama was also hand-held (sort of: I leaned the camera on a hand rail to avoid camera shake) but didn't have the stitching issues you seemed to have. It's a 5 or 6 frame stitch of Salzburg:
http://farm3.static.flickr.com/2420/2321642997_dbc1daff37_b.jpg

[Lloyd Chambers]

I agree it's a great tool and it does an incredible job with even difficult frames that are really badly mismatched in perpective (eg handheld). With even modest attention paid to the entrance pupil (aka "nodal point") the results from CS3 Photomerge are outstanding. Here are a few examples:
http://diglloyd.com/diglloyd/2007-06-blog.html#20070603MorePhotomergePanos

[Ctein]

Dear Bill,

The relationship of image elements front to back is not affected by the focal length of the lens you're using nor whether you are using panoramic software. That is solely a function of the position of the camera relative to the subject.

The difference you are seeing between your panoramics and your wide-angle photographs may have more to do with how the image gets mapped onto the sensor. All camera lenses have to map a spherical field of view into a flat plane. For wide-angle lenses, this can produce some substantial distortions near the edges of the field (the well-known "oval face" problem in group portraits). When you stitch together a series of photographs for panoramas, you're rotating the camera to make each photograph. You are mapping spherical to spherical (roughly) and those edge distortions are much less significant.

Essentially it's operating like a swing-lens panoramic camera instead of an ultra-wide lens panoramic camera (like the Linhoff or Fuji 6x17 film cameras).

Note that Photomerge gives you several options for influencing how the geometry gets mapped (auto, perspective and spherical), and Photoshop/Elements has distortion tools in its filter kit that you could use to emulate the edge distortions of the wide angle lens if you really cared to.

~ pax \ Ctein
[ please excuse any word salad. MacSpeech in training! ]
======================================
-- Ctein's online Gallery http://ctein.com
-- Digital restorations http://photo-repair.com
======================================

[Bill Poole]

Thanks to Ctein for his explanation as to why experience some images as "wide" while others covering the same area seem--for want of a better word--"normal." The basic point I was trying to get at was that digital stitching seems to create a new kind of image rather than duplicating an old kind of image in new way. Thanks again, Bill.

[Dennis]

Bill & Marco: I have to disagree with you on the matter of stitched images looking different from images taken with a WA lens. They might due to geometry, but then pictures taken with different WA lenses vary due to geometry (such as the Tamron 11-18 versus Sigma 10-20). What Bill describes as "other optical characteristics of a near-normal lens--for example the relationship of image elements front to back" is just perspective and is wholly dependent on where you stand. If you tend to shoot from farther away with a tele even though you intend to stitch multiple images, then you'll get a different perspective, but that doesn't make the stitched image inherently different from an image shot with a WA lens; it makes it similar to an image shot with a WA lens from farther away.

Marco, I can understand your point about large reproductions of images shot with a tele looking disconcerting (my word) because such a narrow portion of a normal scene occupies such a large portion of your FOV as you view it, but that has nothing to do with an image stitched together from many shots taken with a tele.

Mike: did you really write "Phun" ?

- Dennis ... waiting for July 1 so I can download a trial version of Elements 6 :) (There's a bug in Adobe downloads - for Windows anyway - that prevents them from starting during the month of June !)

[Hugh Crawford]

The "look" of the different renderings is not perspective but is projection. IE how do you represent something that is not flat on a flat surface. When I hear someone say that a rectilinear wide angle lens has no distortion I always think it is funny that a ball near the edge of the frame in a photo taken with a rectilinear wide angle lens will be very distorted , but the same ball photographed using a lens displaying extreme barrel distortion will appear round.

Most stitching programs will give you a choice of the projection or distortion that they apply.

Here is a better explanation
http://slash72.club.fr/gurl/MODES-PROJECTION/en_prefecture_rectiligne.htm

hugh

[Ctein]

Dear Hugh,

That's a wonderful page of projection examples; thanks!

And the link at the end of it led me to this project which is just wunnerful!

http://www.seamlesscity.com/artwork.html

pax / Ctein

[Roger Whitehead]

Here's a different sort of panorama, in which I walked the length of a long, low building to get the (handheld) shots. It's scruffy but fortunately only an experiment.

If I were to do it again, I'd take more care over keeping my distance from the subject constant. Nonetheless, PE6 has done wonders with the material I gave it: http://www.flickr.com/photos/rogergw/2369830797 .

[Greg Miller]

Those cracks are interpolation artifacts. Photomerge uses the equivalent of a 1 pixel hard edge brush on the layer masks to mask out each layer. If you resize without flattening then Photoshop will add white pixels in the interpolation process since it is not seeing merged layers during the process but only each layer independent of each other.

[Alan George]

All my work is done this way. It kicks the sh@t out of what I can get from my 4x5.

http://alan-george.com

[Bill Poole]

Alan:

Hah! Unless my eyes deceive me, we are doing similar work in many of the same neighborhoods.

http://www.legiblelight.com/2008/05/better-homes-and-autos-ashbury-heights.html

[Stew]

Mike,

I've also had very good results with PS Elements 6 panorama feature, much better than with previous versions. I'm far from a pro but I've gotten what I think are very good results from hand-holding my camera (no tripod needed), simply by concentrating on keeping level and making sure to overlap the shots sufficiently.

I used to use PanoramaMaker3 but now only use PS Elements 6.

For example, this shot was hand-held: http://www.flickr.com/photos/stewstryker/1441955565/

This was from a tripod:

http://www.flickr.com/photos/stewstryker/2386692671/