[New post] Pushing the limits of repositioned panoramic photography
Brian Lawler posted: " In several previous articles I have written about the process of repositioned panoramic photography. It works best when there is no perspective – strictly two-dimensional subject matter. With my experiences with street art (see the most " thelawlers.com
In several previous articles I have written about the process of repositioned panoramic photography. It works best when there is no perspective – strictly two-dimensional subject matter.
With my experiences with street art (see the most recent article here), this works perfectly. Photographing flat art is what works best with this technique.
Photographing three-dimensional subjects causes this technique to fail with near-certainty. I think the first time I tried this seriously was when I attempted to photograph a huge Heidelberg printing press by taking an image, then taking three steps to the right, then taking a photo, etc. The resulting image, after assembly, was a catastrophe.
This is easily explained: perspective does not exist on flat subjects, and real things like printing presses have dimensionality. They have depth, and at every point on the camera plane that perspective is different. There is simply no way for software to reason-away differences in perspective.
So, on a recent visit to Burano, Italy, I attempted to do it again. This time, instead of a 20-meter long printing press, I took a series of repositioned photos of an entire city block along a canal in that charming town, and attempted to put them together into a single repositioned panoramic photo.
I worked, kinda sorta. Well, not very well really.
But, enough of it worked that I persisted, and I was eventually successful in making a single image of the entire block, and it is charming. In the process, I took some liberties with the sides of buildings, the bows of small boats, and the waterway that separated me from the subject matter.
I took a total of 27 photos, each about 10 careful paces apart (I was not carrying my ball of string or my tape measure). The overlap between them was significant, but so was the perspective variance and the distortion of the wide angle lens (Canon 16-35 f2.8 at 21mm).
To stitch them in Adobe Photoshop – which does a spectacular job on repositioned photos taken of two-dimensional subjects – I first had to take some of the distortion out using Photoshop's Perspective controls. I put a baseline guide in, then two vertical guides, and I used an Action to perform the same perspective change on each image, straightening its vertical lines – mostly.
After that I used Photoshop's Automate>Photomerge function, with its Reposition option to put these images together. It didn't do very well at all. I tried several times with subsets of the images, and was eventually able to create two big chunks of the repositioned photo that I then assembled into one using drag-and-drop. This final image was OK as a starting place, but there were fractional boats in the canal, and mooring posts that disappeared half-way to their pinnacles.
It looked pretty good, but it would take too much work to make it into a convincing photo.
So I gave up on Photomerge and decided to build this image by hand, assembling each image adjacent to the next by putting each onto its own layer, then creating a master mask to cut out the perspective parts that don't work. This was complicated by slight differences in size and camera angle, so on each image I did some scaling and a bit of distortion to get the image to fit its neighbor to the left. Then I put the next image in, and did the same until I had completed the block-long photo.
Most of the buildings on the block are conjoined with their neighbors on both sides. There are only a few alleys on the block where you can see into the short distance to another building in back. So, with the exception of the rooftops, which have considerable depth, the façades of the buildings are two-dimensional.
I began my work by using the Magic Wand tool and selecting the sky, then making a drop-out mask with that selection on each image. After I had positioned all the images together I built a master sky mask by assembling all of the separate masks into a new channel mask. I replaced the original cloudy blue sky with a similar cloudy blue sky image I took in 2017.
The most difficult part of my master sky mask was television antennas. On top of almost every building there are skinny poles with various antennas attached. These didn't select well with the Magic Wand, so I enhanced them with the paint brush on the sky mask, allowing them to be seen better against the sky.
I also had to remove parts of buildings in the background and visible from the street where I took the photos. This was done in the sky mask; I replaced these architectural artifacts with sky, and it is very convincing.
When the buildings and the sky finally made sense, I had to work on the boats in the canal. There is usually one boat per building in the photo – sometimes two – and each had reflections on the water of the canal toward the camera. Because of my intervals, the perspective of those boats, which are much closer to the camera than the buildings in the back, changed considerably from photo to photo. To remedy this, I took the best boat photo from each of the two overlapping images, and pasted it on top of the composite photo. This worked well, but I had to invent some of the reflections in the foreground.
When all was complete, I had my repositioned panoramic photo. At final resolution I can print this at about 5 meters long (about 15 feet) by 38 cm. tall (about 15 inches) at 300 ppi resolution. I don't have any plans for the photo at present, but perhaps the city of Burano would like it for their City Hall. Or they could just walk outside and look at the real thing.
No comments:
Post a Comment