A remark on merging large format images.

[Leonard Evens]

I wanted to photograph a building facade which, from the available vantage point, is much too wide for my 75 mm lens to encompass. It is possible to do this using a panoramic head and stitching software, but I decided to try something simpler. I set up the camera to be as level as I could get it and square on the subject. I shifted to the right, made an exposure, and then shifted to the left and made a second exposure. This covered the subject with quite a bit of overlap. My plan was to merge the scanned images in my photoeditor (which happens to be Gimp). But I encountered problems getting the two images to register properly. After some thought, I conjectured that the problem might result from a slight shift of the lens axis between exposures. I then did a careful mathematical analysis which yielded some interesting results.

Suppose you shift the lens axis from the perpendicular by a small angle delta, which is measured in radians. I found that the total change in the length of the image is very slight---approximately proportional to the square of delta. But the change on each side, while small, is larger---approximately proportional to delta. One side gets smaller roughly by this amount and the other side gets larger. When you add the two, they cancer except for a residual term which, as I noted, is proportional roughly to delta squared.

Since when you merge images you try to register opposite sides of the two images, if there is a slight rotation between exposures, you can encounter a significant shift. In my case is amounted to about 80 pixels in 5000 or 1..6 percent. That could be produced by a delta of about one degree.

Note that this sort of problem is unlikely to arise with smaller formats involving lower pixel resolutions, so it seems characteristic, for now at least, of large format photography. I have to think some of possible ways to avoid such slight shifts! If that proves impractical, one way around the merging problem might be to use primarily one image and then merge it with a truncated version of the other, so there is little or no overlap, but I have to think of how best to accomplish that.

[Julian Boulter]

Hi Leonard,

Did you shift using the front standard or rear standard?

I use this technique often to create stiched panoramas in CS2 using my 110XL Super Symmar (which has enough coverage) by making 2 exposures with the rear (film) standard shifted to the left and then to the right. Because only the film moves behind the lens, the images stay in perfect registration and I have no issues in stiching them. The only issue is light fall off at the extremes of the shift.

If you were indeed moving the rear standard could it be that the wider 75mm lens would exhibit the issue you describe more than the longer 110mm lens?

Julian

[Leonard Evens]

Julian,

I am aware that in order to prevent parallax errors, one must move the back. Since there was some parallax error for the facade relative to a distant building, I think I must have moved the front instead. For the plane facade of the building, this shouldn't matter because it was very close to being perpendicular to the lens axis, so the parallax shift should be essentially constant across the face. That must have been my reasoning at the time although I knew about the parallax problem when moving the front. Also, I probably figured that the facade was far enough away that parallax would not be detectable, but I was wrong about that. In any case, even if my reasoning is faulty, I don't believe the parallax shift had anything to do with it. It amounted to at most 15 pixels out of about 5000 and the shift I observed was about 80 pixels. Needless to say when I do this again, I will be much more careful and not take any shortcuts.

It is good to know that you do this successfully on a regular basis, which suggests that it is possible to prevent a shift in the lens axis sufficient to produce a detectable shift if one is careful. What camera do you have? Mine is a Toho FC-45X. It is not the absolutely most rigid camera available. It is good enough for almost everything I do, but it may not be up to this kind of task.

The viewing angle does play a role. I said that the difference is proportional to delta, but the constant of proportionality varies inversely with focal length. So it would be about 50 percent larger for a 75 mm lens than it would be for a 110 mm lens.

There are a couple of other issues that I've considered.

First, one might think that this all depends on where you focus. If you focus at either extreme side rather than at the center, then you would slightly magnify or contract the change in length of the two sides combined. But further analysis shows that the ratio of the lengthening of one side vs-a-vis the shortening of the other remains the same.

Second, while I haven't done the detailed analysis, it seems reasonable that a shift in the angle the back makes with the lens axis should produce an effect very similar to what happens when you shift the lens axis. In principle, you want the back to be perpendicular to the lens axis, but that will never be exactly true. The degree it is off from perpendicular may change. That could happen when moving the back, for example, but it could also happen when inserting the film holders, even if the back is left fixed. It is possible that this was what caused the large shift I observed. It could even be a combination of multiple factors. Another reason for being very careful, but perhaps the answer is to get a new camera. ;-)

[Gary Beasley]

You may have had some inadvertant swing or tripod rotation causing the perspective distortion you describe.

[D. Bryant]

Leonard,

I make stitched compositions quite frequently by shifting the front standard maximally right and left. With my camera I have no rear shift - only rear swing. I'm using Photoshop CS3 to complete the stitching and in my experience it is virtually impossible to have the two exposures match up perfectly. IOW cropping of the image is always required.

Knowing that, I plan my framing accordingly. My experience using a 75 mm S.A. is that there is always image distortion, especially at the edges of the image. I'm certainly not mathematically gifted as you are so I just accept the pit falls of this technique.

I've found that careful subject selection and the use of longer focal length lenses help to minimize the distortions and mis-alignments. My 150 mm lens is my favorite for doing these kind of shots.

Other things to that come in to play is the difficulty of ensuring that the camera is perfectly perpendicular to the main subject plane, and the micro movement of the camera when inserting film holders. Also since I have some amount of play with the film holder in the back of the camera there is virtually no way to have two different exposures align perfectly, not to mention that any two sheets of film will have the same placement from exposure to exposure.

I have also found that frequently the final image after being stitched (and cropped) will have an aspect ratio that is very close to 7x17 or 5x12.

My 2 cents,

Don Bryant

[Julian Boulter]

What camera do you have? Mine is a Toho FC-45X. It is not the absolutely most rigid camera available. It is good enough for almost everything I do, but it may not be up to this kind of task.

I have an Arca Swiss Discovery attached to a Manfrotto geared head on a Gitzo 1348 tripod so it's about a stable a large format platform as you could hope to have, indeed I put this kit together specifically to shoot stiched panoramas. I also use a wide angle bellows to make it easier to get 5cm of rear shift left and right for each shot. The 110XL is perfect for this application.

I prefer the shift mechanism on the Discovery opposed to the 'duck billed' type shift of the F-line model, it just feels more precise to me.

I've never actually tried the method of shifting the lens as I was always put off by the potential parallax problem but I might give it a go now out of interest.

Julian