Large Format for Stitching

<p>If you are goig nto do the 35mm size DSLR mounted on a 4x5 view camera you don't want a Sinar F series camera. You need a Sinar P or C series an Arca-Swiss F or M, or a big Cambo, Toyo, or Linhof. A Linhof Technika should support a medium format digital back but I haven't tried that. I have tried it with a Sinar and it doesn't work. to carry the additional weight. You need a heavy duty rear function carrier. </p>

<p>Darn it Ellis, and I was this close to ordering it......</p>

<p>What is the frame of an F made with? I thought I saw MF digital backs on them before, but I could also be delusional.</p>

<p>How about something like the Toyo View 45 d? Think I might need to go do a little reserach on weights and other setups.</p>

<blockquote>

<p>Bob Salomon - Lastly, what field curvature? Is it in the lens or the technique?</p>

</blockquote>

<p>The panoramic stitching technique produces a naturally curved, or more properly, "faceted", field.</p>

<p>Imagine you have a wide angle lens that will cover a bit more than 90 degrees, so you make a single row panorama moving the camera in 90 degree increments, taking 4 shots. There's interesting foreground as close as 3 feet, interesting background to infinity, so you focus at a hyperfocal distance of 6 feet.</p>

<p>You've just shot 4 sides of a "box". Assuming the plane of focus is reasonably flat, each image covers, at the plane of best focus, 6 feet from the camera, and at the outer edges, it's 6/sin(45 deg) = 8.5 feet away.</p>

<ul>

<li>Shoot a 20 shot panorama, and you've got a 20 sided prism, to all intents and purposes, a cylinder.</li>

<li>Shoot a multi-row panorama, and you've made a polyhedron. Use a large enough number of small angle changes, and you're approximating a section of a sphere.</li>

</ul>

<blockquote>

<p>At the nodal point?</p>

</blockquote>

<p>No, for two reasons. First, no one does panoramas from the "nodal point", that a common misuse of the term. Actual nodal points only matter to macrophotographers. Panoramic stitching is done by moving around the lens's "entrance pupil".</p>

<p>But more importantly, Rodeo Joe described exactly what he was doing. He moved the camera to a new point in front of each store on the street, to put them together into a "perspectiveless" panorama. This is called "orthographic stitching", or usually just shortened to "shooting an ortho".</p>

<p>If the subject is closer then 100 yards or parts of the image are, you want to rotate around what is referred commonly in photography as the Nodal point. If nothing in the scene is within 100 yards the nodal point does not matter. common technique is to overlap each shot about 30% so the stitching software can find common points to stitch in each shot. This is also why the rotation around the nodal poit matters as that eliminates parralax errors between points as the camera is rotated.<br>

Good stitching software will let you select from the same set of images if the final panoramic will be rectalinear or cylindrical or both. Or spherical or all 3.<br>

For maximum detail a longer lens is used rather then shorter lenses and the camera is set to portrait rather then landscape but more images are required and the file sizes can get very large depending on the camera used.</p>

<p>100 yards? Good God, Bob, where did you get such an incorrect rule of thumb? Since my explanation of curvature of field went right by you, I won't bother to work the math to show you how wrong you are. I can recommend some really basic reading, if you like.</p>

<p>As for your comments on the nodal point, as I said before "that a common misuse of the term."</p>

<p>I also have a bit of experience setting up a cirkut camera. For those of you who have never seen on, it's a camera that runs the film past a narrow slit as the camera turns and takes a panorama. The secret to getting it to work properly is to get the camera rotating around the nodal point of the lens. By doing this you eliminate the image skewing. The skewing occurs as a result of having to add the lens movement into the equation. If the lens is moving at all as objects move from the right side of the field to the left, near and far objects will start experiencing stereo separation. When you're trying to synchronize the moving image to the moving film this stereo skew dramatically impacts focus.</p>

<p>If you’re going to use your digital camera with its own lens, here are some things that’ll make your job easier. Make yourself a camera mount that rotates around the nodal point of your lens. That is, keep the nodal point stationary as the camera moves. They are easy to make. For as accurate as it has to be it can be made out of wood or any other convenient material.</p>

<p>Build a fork like the front standard of a view camera, and mount a cradle in it that places the nodal point of lens directly between the pivot points. By swinging it you will tip the camera up and down while keeping the nodal point stable. Place the tripod mounting hole for this setup directly under the nodal point and you can swing the camera from side to side and maintain nodal location. If you can’t find the nodal point specs for your lens, use the location of the shutter, it’s close enough.</p>

<p>Now that you can create images with no stereo skew, limit the field of view of your images. If your depth of field can handle it, use a long lens rather than a short one. (Don’t use a telephoto lens. The nodal points for front and rear elements are not the same point. Both points are outside the lens, one in front and one behind.) A short lens introduces all those crazy distortions that stitching programs are supposed to remove but don’t. You are placing a number of flat images onto the inside of a round sphere. The more images you use to do this the fewer distortions you will have between images. Imagine covering the inside of a ball with six images. You are stretching a cube into the shape of a ball. With six thousand it will look pretty much like a ball before any corrections. If you can't sacrifice the depth of field with a long lens, use a short lens and take a lot more images so that you can use the sweet spot in the center that is relatively free of distortion. Before you start your stitch, crop your images down to this part of the image and about a fifty percent overlap. You will find that your stitches will come out relatively free of distortions and anomalies.</p>

<p>Cirkut camera manual: http://cgi.ebay.com/Cirkut-No-8-Outfit-Instruction-Manual-/220718702788?pt=LH_DefaultDomain_0&hash=item3363dc20c4</p>

<p>Nodal or entrance pupil rotation doesn't eliminate the natural perspective effect you get from changing viewing angle relative to the subject. A rectangle will still come out looking like a banana! Nor does nodal swivel eliminate lens distortion, and it's lens distortion that primarily makes stitching awkward. I'd say that Bob's suggested 30% overlap was a bit optimistic, and a better option would be 50% overlap - 25% per side - with the longer side of the format vertical (portrait orientation). This also maximises the viewing angle of the lens and gives more real coverage for a given focal length.</p>

<p>Right, I think I am going to put this on hold while I figure out the sliding back portion. I have a fairly simple idea, just don't know everything.</p>

<blockquote>

<p>Rodeo Joe - Nodal or entrance pupil rotation doesn't eliminate the natural perspective effect you get from changing viewing angle relative to the subject. A rectangle will still come out looking like a banana!</p>

</blockquote>

<p>That's actually the "projection method". With software stitching, you can select a projection that doesn't distort rectangles.</p>

<ul>

<li>Rectilinear projection renders straight lines, regardless of direction, as straight lines, so rectangles stay rectangles. The only way to get that in "hardware" is with a wide angle lens (being "rectilinear" is the design goal of all wide angle lenses) and a flat piece of film (or what the OP is doing, moving a small sensor around behind a wide angle view camera lens). That's why, to really push this, you need a fairly special camera, like a Fuji 617 or Xpan.</li>

<li>Cylindrical projection, where the image is projected on the inner surface of a cylinder. That's easy to do in hardware: "old fashioned" panorama cameras like a Widelux, Horizont, or Cirkut work that way. Vertical lines stay straight, horizontal lines become banana warped. And, a curve that matched the curve of the film becomes a straight line. That's why you set up the group in a semi-cylinder, so you get a rectangular group portrait (and a banana warped school behind them, like my dad's class photo).</li>

<li>Spherical projection, where everything is projected on the surface of a sphere. You can't easily get spherical film, but you can get spherical lenses (fisheyes, actually easier to make than rectilinear wides) or mirrors to project the spherical image on flat film.</li>

</ul>

<p>The beauty of computer processed images is that you can apply whatever projection method you want, with the laws of physics. Aside from the common rectilinear, cylindrical, and spherical, some also offer Mercator (great circle), constant area spherical (ortho), and obscure projections that even I haven't used.</p>

<blockquote>

<p>The secret to getting it to work properly is to get the camera rotating around the nodal point of the lens.</p>

</blockquote>

<p>Actually, the "secret" to getting a Cirkut working is to get the camera rotating around the "entrance pupil" of the lens.</p>

<p>Any technique that rotates the entire camera, lens, and film as one unit (including a slit-scan like a Cirkut or an old film Seitz Roundshot) needs to rotate around the entrance pupil to eliminate the perspective effect you described. However, there's another optical effect that can occur that leads to a loss of horizontal resolution that can be mitigated by rotating around the "front nodal point" of the lens, so you get the very best results from a Cirkut with a lens that has a "coincident" entrance pupil and front nodal point. But if they're not, the entrance pupil is a higher priority than the front node, for just the reasons you described.</p>

<p>At one point, the most popular panoramic cameras were "swing lens" cameras like the Widelux or Horizont. That's when the term "nodal point" got equated with panoramic photography, because those cameras rotate the lens around the "rear nodal point" while keeping the film stationary. Again, an optimal lens will have the front node as close to coincident to the rear node as possible (for image sharpness) and the entrance pupil also as coincident to the rear node as possible (to reduce skew).</p>

<blockquote>

<p>Make yourself a camera mount that rotates around the nodal point of your lens.</p>

</blockquote>

<p>Entrance pupil, always, for stitched panoramas. Especially since zoom lenses and retrofocus wides are nowhere close to coincident in their entrance pupils and front nodes.</p>

<blockquote>

<p>If you can’t find the nodal point specs for your lens, use the location of the shutter, it’s close enough.</p>

</blockquote>

<p>The OP's camera has a focal plane shutter. Now, the optical location of the aperture, sighting through the front of the lens, instead of looking at diagrams of where it is from the side, isn't just "close enough", it's exactly right, because the entrance pupil is the location of the aperture as seen from the front of the lens.</p>

<p>You don't need to look up an entrance pupil location, though, there are a number of techniques for "dialing in" a camera location, including ones that are based on eliminating perspective skew.</p>

<blockquote>

<p>Don’t use a telephoto lens. The nodal points for front and rear elements are not the same point. Both points are outside the lens, one in front and one behind.</p>

</blockquote>

<p>Which is totally irrelevant, as the <strong>entrance pupil</strong>, the point that actually matters, is still typically inside the lens. I have several telephotos where it's almost perfectly located.</p>

<ul>

<li>The Nikon 135mm f2.0 DC has it a bit behind the lens midpoint, it's 77mm behind the front filter ring of a 120mm long lens (rounded to the nearest mm). It actually puts a lot less strain on a panorama rig than say a Nikon 17-35mm f2.8, which gets set 36mm farther back, and weighs more. It's a brilliantly sharp and well controlled lens for panoramas.</li>

<li>The Nikon 70-210mm f4-5.6 (an oldie but a goodie) has it well placed, inside the lens from the 70-167mm settings. At the 135mm setting, it is almost perfectly balanced on a light SLR like a Nikon D90, no strain at all on the pano rig.</li>

<li>The Nikon 200mm f4 macro has it a few mm forward of the center point at infinity, and it moves toward the rear of the lens as you focus the lens closer (yes, there are such things as macro panoramas). At 2:1, it reaches the rear lens mount.</li>

<li>The 70-200mm f2.8 VR is usable up to about its 150mm setting. The 120mm setting (right between the 105mm and 135mm marks) puts the entrance pupil abeam the center of the tripod foot, so you can actually shoot without the longitudinal rail mounted on the pano rig, which is handy considering the weight of that lens, and it's wonderfully balanced. </li>

<li>Nikon 105mm f2.8 AF micro-Nikkor inside the lens for the entire range from infinity to 1:1.</li>

<li>Nikon 105mm f2.5 Ai-S inside, 10mm forward of the rear of the lens. Another lens that balances nicely on lighter SLRs, eliminating strain on the panorama mount.</li>

</ul>

<p>Actually, if you've got a lightweight panorama rig (Nodal Ninja, Panosaurus, Gigapan, or a lot of home-made rigs), the entrance pupil locations of telephotos like the 135mm f2.0, 105mm f2.5, and 70-210mm f4-5.6 puts a lot less strain on the panorama rig than the settings for a 12-24mm f4, 17-35mm f2.8, 20mm f2.8, or even a 50mm f1.8. Optically and physically, they're nearly ideal lenses.</p>

<p>Which brings us to a little "surprise" you get with telephotos. The Nikon 70-200mm f2.8 VR has its entrance pupil line up with the center of the tripod foot at about the 120mm setting (you can skew it and fine tune the zoom setting). That means that you can just drop the lens onto any standard head that can pan (most ball heads, all 3 way heads), level it, and shoot a parallax free panorama, without exotic equipment. And it's got a rotating tripod collar, so you can do verticals.</p>

<p>The old 70-210mm f4-5.6, or newer 70-300mm f4-5.6 or f4-6.3 lenses all have a "sweet spot) where you can rotate around the camera's tripod mount. It's usually around 200mm, where the entrance pupil has moved outside the lens rear and into the camera.</p>

<p>Although there are some telephotos that can't be used. My 300mm f2.8 AF-I cannot be used for panoramas. It has a nearly telecentric front stop, not just outside the lens, but 356mm behind the camera's film plane. But it's an exception, virtually all telephotos are excellent performers for panoramas.</p>

<p>OK, I've been inexcusably rude and made 5 posts to other people, while completely ignoring the OP.</p>

<blockquote>

<p>Zach Ritter - So what is the general thought on using large format with small format (i.e. Nikon) cameras for stitching?</p>

</blockquote>

<p>it's not hard to get up and running with a DSLR improvised on a LF camera. But, as you're hearing, it is hard to get it going in a way that gets you the best results.</p>

<p>On the plus side...</p>

<ul>

<li>Shooting all your tiles through the same, immovable lens completely eliminates all perspective shift or skew problems.</li>

<li>Shooting through the same lens and not changing direction eliminates changes in flare. This is true of both using shift lenses like Ellis and using an SLR on a view camera. Conventional panorama rigs frequently have a change in image character as you pan or tilt: veiling flare alters the contrast and color of every frame, and ghosts and "blobby" flare appear in different spots in each frame.</li>

</ul>

<p>On the minus side</p>

<ul>

<li>Mechanical stability is a problem. A view camera rear standard is built to shift and tilt a lightweight film holder or ground glass. When you put a pound or three (500-1400g) of DSLR on it, you have the rear standard tilting when you raise or lower the camera, and the whole body twisting when you shift. Generally, the stitching software will compensate for some of it, but if a rear standard or rail actually bends a bit when you pull the ground glass and drop in your camera carrier, you get a focus error.</li>

<li>Still on stability, the average view camera is not built to contend with the focal plane shutter and mirror of a DSLR. Rigging to do something like firing the DSLR shutter on a 4 second exposure, and firing the view camera shutter 3 seconds into the exposure is possible, but requires an electrical shutter release on the view camera. And, the DSLR gets noisy on 4 second exposures, even if there's only actually light for 1/15 second of it.</li>

<li>As others pointed out, APS DSLRs outresolve view camera lenses, even the newer "digital" ones.</li>

</ul>

<p>Here's a couple of helpful tips...</p>

<ul>

<li>If you don't mind making a lot of shots, stick with an APS DSLR. FF DSLRs run into problems that others described earlier, the lens mount occludes the sensor with shorter lenses at larger shifts.</li>

<li>Buy or make a "bag bellows", the DSLR mount tends to put the camera's plane of focus pretty far behind the normal film holder plane of focus, so you end up moving the standards closer together to compensate. Since you're using wide angle lenses, this can compress the bellows to the point where movements are restricted. A "bag" lets you move freer.</li>

</ul>

<blockquote>

<p>What is the frame of an F made with? I thought I saw MF digital backs on them before, but I could also be delusional.</p>

</blockquote>

<p>An MF back doesn't have a mirror or shutter, is lighter than an SLR, and usually is used for a single shot, not tiling, so there's no need to worry about stuff moving between shots. Oh, and you made it an "either/or" proposition. Having seen something doesn't impact your ability to be delusional. It's like the old saying "just because you're paranoid doesn't mean someone isn't really out to get you".</p>

<p>OK, back to stitching.</p>

<blockquote>

<p>And I would think, that as long as you are turning the front of the lens, you are altering the perspective and creating a "curvature". However, most software can correct the parallel line issue, so just strike that one to my own stupidity. Really, it is those movements that I am after.</p>

</blockquote>

<p>I simulate all movements. If the scene exceeds the DOF capability of my lenses at the resolution I want, I use focus stacking, then feed the stack into a panorama stitcher. If you use the entire focus range of a scene for every shot, this can take forever, but it gets a lot easier if you stack only the focus range of each tile. I've done some pretty weird stuff that way, out to infinity above the horizon, and 9 inches from the camera at the bottom of the image. Don't try this with PhotoShop's stitching capability, do the focus stacks in CombineZ, the stitching in something like Hugin or PTgui. The nice thing about a focus stack is that it's true deep DOF, not just a tilted plane of focus like you get from tilt or swing movements.</p>

<p>It can also get around curvature (or faceting) of field.</p>

<blockquote>

<p>I already have one of the older single row pano heads from Manfrotto, and it works great with longer lenses, but I don't like the distorted feel I get with shorter lenses. I don't know the right technical terms for this situation, it just doesn't feel right.</p>

</blockquote>

<p>That's where you need to read up on the different projection modes, like I wrote about earlier. You'll often find different images look better in different modes. But remember, one reason why these things might not look right is that stitches from short lenses go beyond the capability of the human visual system. There's no way to put a 360 degree panorama on the screen, or print it, that will look "right" to a human, unless you mount it in a panopticon (a wrap around presentation). I've done that before, it's expensive, and it requires a lot of exhibition place. I printed one 4 ft x 25 ft, and mounted it in an 8 foot diameter cylindrical panopticon once.</p>

<p>Hi,<br>

This has been a very useful thread, including the discussion about the general challenges making panoramas with DSLRs. I agree with Joseph about the difficulties of attaching a DSLR to a 4x5. Certainly something like the Nikon D3 or D2x would introduce problems because of weight, as well as some other technical issues (e.g., resolution, CA from film lenses). However, I do think it is doable with lighter weight DSLR or even better the new mirrorless like the Sony NEX. With the right mount, the load is less than the ground-glass back with a film insert, and the center of gravity is better located with the NEX. So, I think that the rear standard can handle these attachments. I have done several large panos using a Toyo 45G, both with a D70 and NEX 5. Below are links to examples. Even thought these are cropped, they are quite large.<br>

With the D70

With the NEX 5

Good luck.</p>

<p>Joseph, thank you you for the wonderful input. It's nice to see someone well grounded in the present century (I'm not) providing up to date input. I will concede on the entrance pupil vs nodal point because I don't remember my sources to be able to challenge it. In the end, what's important is to improve your camera placement if you're making a number of images to stitch by moving your camera around in the scene. Rotating the camera around a somewhat stationary point around the lens will produce a better set of image files to work with than snapping shots using your tripod movements.</p>

<p>Rodeo, I didn't say the nodal pivot point eliminated lens distortion. I recommended eliminating stereo skewing by reducing lens movement while taking several photos to stitch. I also recommended using the center portion of the image where there is less distortion rather than using a wide angle lens. Using a smaller field of view avoids the problems you get when you try to stitch wide angle images together. Try stitching 4 Wide angle images of a square scene together. Then do the same with 25 images and see how much better it turns out. Unless there is something complicated going on in your scene, it will probably turn out without anomalies using the 25 images. Yes there is still distortion from one shot to the next, but the stitching programs do a nice job of working it out if you don't bury them with overlapping wide angle images. There is very little distortion for the program to deal with if you limit yourself to about 20 degrees of image and a 25% overlap on each side by lens selection or cropping. There won't be much to work out between most images, and 25 shots will give you a nice 100 x 100 degree field of view.</p>

<p>There will always be something unnatural about a scene that takes in more than 50 degrees of viewing field. 30 degrees is the sweet viewing field of the human eye. If we want to see past this width we move our eyes to a new sweet spot. Most of this is due to the poor quality of the image on the retina due to it's shape and orientation of the lens. Then, you have competing images from two eyes. Allowing both eyes to take in a 100 degree field of view all at once and getting it to look good is a subject, technique, composition balancing act that is a challenge to any photographer.</p>

<p>This has been a very informative thread.</p>

I just thought about the new mirrorless from Nikon with a very shallow lens mount. It would be good for this right?