In this second episode of a 3-part article on creating photo panoramas, I am focusing on stitching techniques, i.e. techniques involving multiple frames assembled into one. This can be done automatically in camera or manually in post-processing software.
Part 1 (techniques involving a single frame) is online here.
Welcome to the lazy world of in-camera panoramic photography 😉
We’ve all done it. Sweep your digital camera or smartphone in a vaguely horizontal fashion and let the brain in the camera work-out the alignment, stitching and exposure compensation between frames.
If your final display of these panoramas is going to be the web, I’d advocate the use of a Smartphone. Quality is easily good enough, ergonomics often excellent and the range of native sharing features much more fun than those of today’s “connected” cameras.
The clickable picture above, made by my son, demonstrates the pleasing results that can be obtained by someone with no interest in photography (in spite of a few stitching glitches and visible exposure steps). Plus, for the ultimate in immersive Internet sharing, you can go beyond the panoramic format and indulge in photosheres. Rather fugly to look at as a jpg file, they’re a lot of fun to browse with dedicated players. Google have even posted instructions on how to use them in your blog, if you wish to.
Most mirrorless cameras also include a native panoramic mode that will give you similar results in even larger files, so you can realistically think of printing your best efforts. The 4 examples below were created using the glorious Sony NEX-5n (and successors) and previously posted on this page of panoramic Perth and Hong-Kong.
Creating panoramas in this way can be addictive, it is so quick and easy. But the technique has its limitations.
Even the more expensive cameras cannot do away with the inevitable foibles of moving objects, as the two examples below demonstrate. The first was done on purpose as an artistic study on movement. But the second was a totally unexpected mess.
And, secondly, the camera makes all the decisions for you so the final geometry may not quite be what you had hoped for. The fence behind the Moreton Bay Fig tree, above, couls have been made to look straight in dedicated software, whereas the camera chose a projection mode that resulted in this cylindrical distortion.
Still, if you’re happy to live with a few failures and great ease of production, the in-camera method has its merits. It’s the cheapest option here as you probably already own a Smartphone (there are apparently more in circulation than there are toothbrushes …) and/or small camera with this ability. The best in class are remarkably good at dealing with illumination variations between areas, a *very* desirable feature if you’re not willing to spend much time on your pano making.
Two other limitations worth mentioning:
And one major problem : parallax. As your arms swing around the 90°+ arc required to create your panorama, the lens and sensor move about 1 meter from one point to the other. That’s not an issue when there are no objects close by in the frame. But the difference in angle for an object close to the camera relative to the background will make it very difficult to avoid double images. Although not visible at this scale, the Boranup forest image above suffers from this issue in the foreground ferns.
To mitigate this, try to swivel the camera around its axis rather than rotate your whole body with camera at arm’s length.
Enter the world of huge mega-pixel counts, headaches and dreams.
Until I read this sobering article (automatic Google translation from French), my dream camera was the Arca-Swiss Rm3di (or possibly Alpa 12STC). The workmanship and file quality of these two gems is unparalleled. Seeing is believing. But anyone willing to put up with the UX cr@p described in the article is a far braver man than I! A shame really, because I cannot think of a more pleasant way to bankrupt my family than the 12STC, a Phase One IQ2 260 achromatic and a slew of Rodenstock HR Digaron-W lenses. Ah well, saved by lousy ergonomics!
It’s too bad because these technical cameras allow handy shifts of their digital backs that allow the photographer to capture multiple areas within the limits of lens coverage. Take 3 shots, one shifted to the left, one in the center and one shifted to the right and you can stitch them very simply into a very high quality panorama.
The RhinoCAM basically holds your lens steady on a tripod base and allows you to slide the camera both vertically and horizontally. It lets you mount medium format cameras which have a large image circle (the disk surface on which the scene’s image is projected) and allows the smaller sensor at the back to capture the large image in smaller chunks equal to the surface of the sensor. It’s a bit like taking a huge image and transforming it into a jigsaw puzzle made only of rectangles.
The RhinoCAM’s claim to fame is it’s ability to transform an Sony NEX-7 into a 150 mega-pixel monster: you need to stitch the 8 individual captures into a large single image, of course. But since each is a rectangular extract from the same image (from the same lens, that doesn’t move) there is absolutely no difficulty in realigning them back into the complete large format image. Move simply from left to right (or up and down) instead of covering a whole grid and you have horizontal (resp. vertical) panoramas.
The Mirex works the same but it’s the lens that moves about, not the back. While that could cause parallax problems with really close objects, it’s a total non-issue in usual panorama conditions. Build is better than the RhinoCAM and the cost is significantly higher. It is very easy to use for panoramas although making a complete 4:3 image is much less intuitive than with the RhinoCAM. It also offers tilts and that’s a big deal.
These differences aside, both work like the technical camera, letting you capture a large image in smaller bits that you reassemble in Photoshop or similar software by simply placing the images next to one another with no other manipulation. In the 2 plane-jane examples above, the lens used is a Mamiya 645 45mm/2.8 (equivalent to a 35mm on 24×36). I have left the edge jaggies and exposures inequalities straight out of Photoshop’s “reposition” Photomerge to show that 3 frames were used for each and give an idea of the final angle of view.
Are they worth it ?
In my mind, not the RhinoCAM. My (limited) experience with AutoPano-Pro tells me that careful shooting using quality lenses always results in excellent final files, making the added complexity of an adapter a bit pointless, particularly with medium format lenses that cannot match the pixel level resolution of modern designs such as the FE55ZA.
The Mirex is another story! It’s build quality is superb and the tilt feature makes it easy to control the plane of focus (see below) and create panos that are sharp from your toes to infinity without resorting to focus stacking or extremely low apertures and their diffraction issues. With a high dynamic range camera such as the Sony A7r, three side-by-side photographs are all you need for very high quality prints up to 40″ – 50″ wide (imagine the complexity if you needed 3 photos for coverage, 3 of each for depth of field and 3 more of each for HDR!). I use mine with very cheap and really-quite-good Mamiya 645 lenses and the results are very pleasing.
In spite of this, I probably won’t be keeping the Mirex and lenses because of my preference for technique number 1.
Shift lenses, such as the Canon TS-E 24mm f/3.5L II work exactly in the same way. Their optical design is basically that of a medium format lens and their coverage is far greater than the surface of the sensor. And their shift mechanism allows you to take multiple frames (fully extended to the left, center, fully extended to the right) and assemble the frames into a single pano.
The Mirex is essentially the tilt-shift mechanism you find in the Canon TS-E 24 and has the added benefit of allowing you to transform many standard lenses into tilt-shift lenses. The Canon, however, has the advantage of a much wider angle (24mm focal length, whereas the Mirex cannot use lenses shorter than 35mm).
This is the way we’re all supposed to do it.
The essential difference with technique 4 is that, instead of taking several pictures of the non-moving image projected by a single lens, we are making several pictures of the moving images projected by the lens as you point the camera in different directions. In technique 4, the camera does not move around. In technique 5, it does. Because each individual frames has distortion (from the lens) and misalignment due to imperfect panning, the software doing the stitching had to digitally alter the images to make them fit. This time, the jigsaw is no longer made of rectangular pieces. The computer has to work out the shapes that permit the best overlay.
So : carefully capture well aligned images using a low distortion lens, manual focus (set where you want focus), manual white balance (or set to a constant position such as “sunlight”) and manual exposure (set to best preserve the global highlights of the scene and identical for all frames), make sure each overlaps the previous by 30% then assemble the resulting files in dedicated software.
In this example, I skipped the careful part and headed straight for easy. As in automatic everything and handheld shot. It shows in the first picture where I hand aligned the individual frames. But the final result is perfect, showing how today’s excellent software allows a bit of latitude in technical thoroughness. Note that the high file quality of the camera used (Nikon D800e) made my exposure sloppiness acceptable, but I am not advocating this. It is always best to start with well aligned and identically exposed frames.
A first investment you can think of is a good panning platform. Once level (most include a bubble level), you simply pan around the central axis of the camera and take your individual shots. Mine is an Arca-Swiss P-zero monoball. It’s graduated, which lets me shoot at very regular intervals and buttery smooth yet very, very stable. Highly recommended and one of the most affordable goodies from that brand. It’s also small and lightweight (see below).
The second investment you’ll require is stitching software. Photoshop handles panoramas, but comes with a learning curve and a rental policy that not everyone agrees with. Plus, it makes no sense to get into Photoshop with that use in mind.
AutoPano Pro is a much more interesting alternative. Not only does it create remarkable images very quickly, it also comes loaded with features for detecting potential panoramas automatically and many other conveniences. You can download a free and full-featured evaluation copy of this software. It will add AUTOPANO watermarks everywhere on your images until you buy it (see B&W sample above).
This technique requires much more work and effort, but will result in huge files of very good quality. The monoball costs about 250€ and AutoPano Pro adds another 120€. A good fixed lens (low distortion and high corner quality at mid-apertures) is also a worthwhile investment if you’re serious about panoramas. Since you will require a medium wide to normal focal length, it will be one you use regularly, not a specialty item.
You can also mix and match techniques. For instance, the picture above shows a photomerge (in Photoshop) that uses two in-camera panoramas made using the Sony A7r and FE 35/2.8 ZA lens. The area to cover was huge so I experimented with two panos. One for the top part, one for the bottom. Since the camera did all of the geometric corrections, merging the two is very easy.
As you can see in the final file, they are not properly aligned because I didn’t start-stop at the exact same spots. And, more importantly, the jpeg output has made it impossible to reclaim highlight detail in the sky. Still, in some situations (low global contrast), this is a worthwhile and very easy technique that produces a very decent-sized file.
In part 3, I will be presenting more “exotic” (as in amusing) techniques for creating photo panoramas.