Stitching as opposed to shooting larger format, revisited

Discussion in 'Digital Darkroom' started by tdub, Mar 31, 2011.

  1. Before I plunk $ for a pano head, I need to know whether I can completely eliminate warp in stitched architectural interior shots (e.g., a living room scene). I was told on another forum that you'll always get some residual warping, that can't be removed. Edward Ingold, in a thread here last Nov stated: "Distortion is removed by converting each image to a spherical projection, blending, then reverting the assembled results to a rectilinear image. That's pretty straightforward in PTGui, especially if you determine the nodal point accurately." This sounds to me that, doing it correctly, you actually may be able to remove all the warping. Is this true? Like the original poster, I want to stitch frames together to get higher res with this type of subject, but can't have any warping. (Assume rotating on NPP axis.)
  2. While you might remove warping, you will not remove perspective shifts from moving the lens. So, if you want a more "true" looking image, you either need to use large format or use a stitching system based off of a large format. I started a thread on building a system based off of large format a few weeks ago. There was a bit of a lively debate about the whole perspective *X&$#**X&$#**X&$#**X&$#* thing. Well worth the read.
    In the end, I am on hold until I come up with a system to move the camera around the back without moving the back itself. I may actually build it right into a field camera. Although, composing could become an issue.
  3. What is the widest angle lens you have right now and what size camera format?
  4. the whole perspective *X&$#**X&$#**X&$#**X&$#* thing​
    No wonder the discussion was lively. I'm not familiar with this technique myself. JR
  5. duplicate deleted.
  6. I have had success using lenses with shift capability, particularly the Canon TS-E lenses. I shift the lens outward and make an exposure at click stop. Using PTGui (I recently upgraded to the most current version PTGui Pro) for sttchign the iamges to gether I export as a "Blended + Layers" PSD.
    I then open the image in Photoshop and if there are any glitches find the best layer that addresses the issue, and edit that layer's mask by painting with white to resolve the problem. The photo accompanying this post is an example of that approach. There were two glitches in the lower right corner tha tneeded to be resolved.
    I also have a full Really Right Stuff panoramic rig ( A PC-1 rotating clamp, CB-10 bar, PG-02VA and clamp, and MPR-192 "nodal slide") . When you start trying to take in an extremely wide angle of view ,defining extreme as over 110˚. of a three dimensional scene and reduce in down to a two dimensional form there are going to be distortions that result from the resulting geometry of rendering such an extremely wide view in 2D . You just have to decide which type of distortion works best or least offensively for the photo you are making.
    I had not seen Edward's technique discussed before and will try it. Do you have a link to that discussion?
  7. this is a view from a single frame that was used to create the above composite. Yes it renders perspective differently that the final composite, but keep in mind that it is a much narrower angle of view.
  8. "I had not seen Edward's technique discussed before and will try it. Do you have a link to that discussion?"
    Ans.: ref post is last on this page:
    "What is the widest angle lens you have right now and what size camera format?"
    Ans: FF Dslr, 20mm lens.
  9. PTgui will take out any warping and lens distortions. Once you have your photos to stitch you can choose what direction you want the image viewed from. You can get the same corrections that a view camera does when it shift the lens.
    So take this for a case
    That is an exterm wide angle view, but I can stitch it any number of ways, as follows
    Those are all made up from the same soure images, just stitched pointing diferent angles
  10. Would you guys still buy PTGui Pro over the built in CS5 ability?
  11. PhotoMerge in Photoshop CS5 is very good. But primarily I use PTGui Pro. That doesn't mean you should or that PhotoMerge is inadequate for the stitching you will be doing.
  12. Thanks, Ellis. I used to use third party but CS5 has been fantastic. So much so, that I haven't even used my kaidan head for sometime now.
  13. What type of subjects and what kind of light ( dim, mid day, etc.)?
    I find that the more spatially elaborate the subject is in term s of near far relationships, the greater depth of field , the greater the depth of field required, the more stringent the need for keeping things level, and the larger the ultimate usage might be, the heavier duty the tools (cameras, lenses, hardware and software) are required. There are some pansI'll make with the camera handheld as well but for most of the work I am doing I am glad to have a good kit of tools at hand.
  14. Is anyone using the Gigapan system? If so how does it rate with other pano-stitching methods and software.
    I have access to one but unfortunately the weather has prevented me from testing with it.
    FWIW, I will be using a Canon G10 but the DSLR Gigapan models also interest me.
  15. Ellis, nothing specific these days and usually always while I'm just out snapping for snaps sake and finding myself making a pano or two in hand-held light. I'm really pleased with the results just from spinning on my heals so to speak. But years ago, we used to make vrml's and also used autostitch (i think it was called?) and PTGui. I still have the head but haven't used it in ages, since like CS3 era.
  16. To echo donald, I'm looking at getting panopro giga also. Does anyone have any insight into how it stacks up?
  17. Is anyone using the Gigapan system? If so how does it rate with other pano-stitching methods and software.​
    I've been using a Gigapan for a few years now (I enrolled in their beta program early on.) I can say only good things about the robotics approach. It enables a far, far superior image capture workflow than what is practical with any manual pano head.
    At each slew position, I usually set the camera to take around 12 exposures spaced 1/2 or 2/3 stops apart. The pano circuit often is set for greater than 180deg (sometimes a full 360deg) field of view.
    The point is to enable maximum freedom in choice of exposure and composition by deferring those decisions all the way to post. The typical print for me thus usually arises out of a subset of the slew circuit, and blends of a portion of each exposure stack.
    The ideal camera for the Gigapan is actually a high pixel count compact digicam, not a DSLR. Exposure blending makes non-issues of the noisier and narrower dynamic range handicaps inherent of the smaller digicam imager. The gain of a smaller, lighter, more portable photography kit is a significant advantage.
  18. No wonder the discussion was lively. I'm not familiar with this technique myself. JR​
    Jeez Jeremy, so I missed an "F" key somewhere.....
    Scott, I think Pano heads+ ptgui works great on single rows, but as soon as you start multi rows you start to see perspective shifts. There is just no way around it because the source images are collecting a different perspective.
    That all said, Single row + WA would probably work just fine for the OP. Sounds like he is going for illustrative and wide rather than massive print.
  19. ... but as soon as you start multi rows you start to see perspective shifts.​
    Perspective shift? Do you mean perspective distortion, the kind commonly compensated for with movements with a large format camera?
    In general, this compensation is really easy to replicate in post (via essentially the same transforms actually.) However, a side-effect of the digital domain warping operation is that it spatially compresses and expands local areas of the image. The potential problem then with single frame captures is running out of pixels, i.e., lose too much detail in sections of aggressive expansion.
    This highlights a key advantage of working with high resolution composites (and by implication, the robotic mounts that make the workflow practical.) If for example you have enough pixels to place 1200 real dpi onto a 16x20 print, what it really says then is that you can afford to lose up to 900 of those dpi to filtering operations, and still get a tack sharp super-detailed print.
  20. Now here's the perspective "corrected" version.
  21. Nope, I am talking about actual perspective shifts that happen from the movement of the front element. Take a window, for instance. From one angle, you can see the left side of it's casing, but not the right. As you move your lens, you can now see both the left and right, and then finally just the right. The severity of this increases with the wider of an arc you make. 90% of people, I would guess, would never notice this, but it bugs me to a degree.
  22. Zach, the camera is rotated around its entrance pupil, so there is no shift of perspective as you describe it.
    I do multi row stitching all the time, typically 3 row x 8 to 10 columns, there is no problem with perspective at all.
  23. Scott, it would be impossible not to have it. You are using a curved field of capture (akin to a fisheye and some ultrawides) vs a flat plane of capture. Not to mention the total size of the arc is going to determine how much you see the effect. a 3x8 with a 200mm will show it much less than a 3x8 with a 14mm. If it didn't have this effect, it would be 100% impossible to do a 360 panorama, or even a 180.
  24. Zach, you really don't konw what you are talking about here, sorry.
  25. Scott, It is elementary physics. Sorry, but if you were looking straight forward, with a single plane, you can't see what is to your side. You have to turn to see that. You are changing your entire viewpoint. Even if your entrance pupil is constant, you are changing where/how you interact with that pupil. To not have this distortion, both the entrance pupil and the front element would have to stay constant since they control the projection to the film plane.
    Maybe perspective is the wrong term, but is still a distortion.
  26. Zach,
    You are confusing several things. If you turn your head the effect can be bad, if you set up a camera on a tripod then the effect can also be bad, but there is on three dimensional point inside the lens where it projects without distortion anywhere, that is commonly referred to as the nodal point but in truth is the entry point.
    To set your camera up so that it projects an error free multi-row stitch you have set the camera and lens to pivot around the center of the entry point.
    This is the best way I have found to do it.
  27. it isn' perspective distortion - i nfact it is the opposite - it is the correct rendering of an extremely wide angle view. If you made you print large enough that you could wrap it around you , you wouldn't notice any distortion as you turned your head from side to side or looked up and down.
    So if the final stitched image does distort the perspective, what is going on? As I wrote before (and was ignored) what you are seeing is the effect of compressing an extremely wide angle of view down to a flat document that occupies a narrower angle of view.
    The best solution must always be a compromise of the geometry, but subjectively you need to start with choosing where you position the camera in relation to the subject to make the most acceptable (to you) two dimansional rendering of a volume of three dimensional space.
  28. Ellis,
    I would never ignore you! I thought Zach was confusing his pallax with his projection. I took your point of representing a three dimensional space on a two dimensional plane as being so obvious as to not cause confusion.
    What I was very interested in in this thread though was your use of the 17mm TS-E and the stitch you have that you said you had issues with in the lower right hand corner. I was intrigued as to what the issues were, I have the 17 TS-E but so far have not had such alignment issues. I use CS4 for auto aligning and so far have been very happy with it.
  29. Scott F, I understand the whole parallax issue (even if I can't spell it). What I can't remember is if entrance pupil and the nodal point are about the same thing. If I remember correctly you want to pivot around the point where all of the beams of light converge inside the lens.
    And Ellis is 100% correct. I was just using the wrong terms. Lets just say when I am in the middle of a day of coding it gets hard to switch my brain to the correct terms for photos. Add to that, that I have had an added emphasis on video lately, and my mind is swirling with about three different sets of nomenclatures. It hurts.
    The problem for me, is if you use too wide of angle, it creates a sense of vertigo. Not pleasant. It is the same unpleasant feeling I get from fisheye lenses now. I used to love them, but now they just make me uneasy.
  30. The problem for me, is if you use too wide of angle, it creates a sense of vertigo.​
    Some kind of rectilinear projection is the default on many pano programs. It works great for FOV of 45deg or less. Try different projections for wider scenes.
    If you have the pano head properly configured for the particular camera, there will be no parallax errors as the camera slews on either alt/az axis. The better pano programs will in addition calibrate out various lens distortions before stitching. The default expectation these days should be of entirely seamless composites, ones indistinguishable from a single shot capture.
    By the way, try Hugin. It's open source, free to use, and excellently implemented.
  31. I have had horrible luck stitching exterior shots of buildings in PS CS5. They always converge toward a vanishing point on either the left or the right side. It looks as though my shooting position was to the side of center despite the fact that I was actually right in front of the structure's center. So bizarre! I tried every option in Photo Merge but to no avail. I've all but given up on the technique.
  32. Zach, Entrance pupil and nodal point are not the same thing. In some lenses, these points are separated by tens of cm. Just browse this table of lens parameters for example:
    It is the entrance pupil location that is the correct point about which to rotate the camera to avoid parallax. All the common methods of finding the no parallax point actually find the position of the entrance pupil rather than either of the two nodal points. The fact that the entrance pupil can be seen simply by looking into the lens makes it easy to find the no parallax point for a single lens reflex camera. See:
  33. I have been using a MultiStitch plate to shoot interiors with both my DSLR (a Canon 5D) and an Imacon 16MP digital back on my Cambo 4X5 Ultima-D. I also use it for product work in the studio. The MultiStitch works by moving only the capture device so there is no distortion in the images for stitching.
    The lens AND the "film plane" remain stationary throughout the exposure sequence. Further, the plate allows for a 2-over-2 pattern and not just a wide panorama. I have used it for shooting up to 2 shot high by 4 shot wide interiors with great results. Sample images can be seen here: Multistitch sample images
    Full disclosure: It is my project, developed over the past several years and I have a patent pending. Please let me know what you think as I am currently bringing it to market.
  34. Full disclosure: It is my project, developed over the past several years ...​
    I've been aware of the CameraFusion product that's been available for a few years now; Kapture Group also has something similar. How does MultiStitch differ?
  35. I've been aware of the CameraFusion product that's been available for a few years now; Kapture Group also has something similar. How does MultiStitch differ?​
    Hi Robert, The advantages are fully explained on the website at but I'll summarize...
    The Kapture Group (K) and Camera Fusion (F) devices are for medium format backs and DSLRs, respectively. MultiStitch (M) is available for either one, and at a cost that is roughly half that of K & F, (even with K currently on sale for 40% off).
    M is SIMPLE both to learn and use and can operate in either vertical or horizontal aspects as easily as turning your groundglass. It has capture chip overlaps built-in and is the size and shape of a triple-thick lens board. M does not interfere with camera movements at all in either aspect. K allows for turning the MF back and uses cranks to move it around with many loose and moving parts to swap out and adjust. F lets you turn the DSLR but then you have to do the math using a sliding scale to position it to get proper overlaps. Both K and F will interfere with camera tilts because they are wider than the camera itself by a wide margin. I do not think F can be used at all in the vertical aspect.
    M does not add so much thickness to the rear of the camera to prevent the use of wide angle lenses with MF digital backs. I even shoot with my Schneider 47XL. K limits you to 75mm, or 90mm if you need movements, and up. F is roughly comparable for focal length with M as both are stuck with the depth of a DSLR mirror box.
    MultiStitch is designed for Professional shooters that want to get back to the photographic advantages of a large format camera while giving clients the digital files they need. It is also designed to extend the ROI of digital equipment that is nearing the outer edge of it's useful life but is still perfectly operational. I have a 16MP Imacon back that I routinely use to produce 54MP image files. I believe you should not have to replace a 5-year-old, $25K piece of equipment that works like new.
  36. "It is my project, developed over the past several years and I have a patent pending. Please let me know what you think
    as I am currently bringing it to market."

    This is a major no no under the user terms and conditions of this site.
  37. Hello Scott, I posted that as a good-faith statement. I am very interested in the photographic process, and I did not want to be perceived as "using" the forum for sales. The full text of what you cut and pasted was "FULL DISCLOSURE: It is my project, ....." I was being honest in my participation, and my second post was in direct reply to a member's question.
    I will read the T&C more thoroughly, although asking for feedback from fellow enthusiasts seems harmless enough.
    Best regards, -Myko
  38. "...This is a major no no..."
    Maybe, maybe not. Myko's first post in this thread was no more than spam-like than Henry Posner's are - factual, helpful, and containing a disclaimer.
    His more recent post does show many more details (including prices), but (a) Robert Lee explicitly asked how his product differed from other units, and (b) I (and I suspect many other lurkers) learned a lot about this field from the information in that post. I don't know Myko, have no relationship to his company, have never tried any of these products, but have an interest in this topic, so I would urge the moderator(s) to cut him some slack.
    Just my $0.02,
    Tom M
  39. Tom,

    We are all welcome to our 2 cents, but when straight forward adverts appear, along with prices and lifted images, I don't
    think I am too far off the mark.

    I too have more than a passing interest in the thread, which is why I have notifications for it, I stitch regularly and use a
    variety of techniques, including keeping the lens stationary and moving the body. Something I have achieved up to now
    with Canon shift lenses and rails. I am also perfectly capable of following a link, and paying considerably less to get a
    board engineered to move a body on should I decide I need to use my medium format lenses and bellows and I can't be
    bothered to make one myself. I assumed most of my fellow followers were of at least similar mental acuity.
  40. Hi Scott,
    ...moving back to the thread topic without hard feelings I hope... I'm curious about your rails setup to work with Canon shift lenses to keep the lens stationary and move the body. Also, I'm curious as to what medium format lens/ bellows setup you use. My medium format lenses (mamiya and hassy) don't have a large enough image circle for stitching more than 2 DSLR frames unless I'm doing closeup work, that's why I ended up on a large format platform.
  41. Mkyo,

    Absolutely no hard feelings at all, I bare no animosity or ill will at all. I hope your commercial venture is very successful.

    With regards the Canon lens rail, I use a very simple arrangement of a RRS MPR-192 rail with a set of mini clamps,
    mainly because I have them not because they are particularly special, a geared macro rail would work even better. I set
    the first shot up with the rail across the direction of shift, either vertical or horizontal, I then take that shot, shift the lens,
    then shift the camera in the opposite direction, this effectively leaves the lens stationary and just moves the back. With
    the new TS-E lenses the image circle gives you a variety of effective sensor sizes, 24x60, 36x48, 36x36 etc etc. so far I
    haven't needed more than an effective 48 mp multi-aspect ratio sensor.

    With regards the medium format set up, so far I have had no need to digitize it. But I use Mamiya 6x4.5, 6x6, 6x7, and
    6x9 backs so have a very large image circle to work with if I ever have the need.

    So far, for my work, I haven't required more than 48mp worth of data.

Share This Page