DoF and Perspective Revisited

[Dick Roadnight]

...but the theory is that the stitching software e-projects the images to try to correct the perspective.

Do you have any concrete comparison showing that the leading softwares are unable to do it? I believe that my post above in this thread clearly shows that there is at least one example where they do an excellent job.

My question is geniune, I have never seen such a case, but it could very well exist.

Cheers,
Bernard

Bernard...

I am not trying to pick an argument with you or anyone.

I agree that your post above in this thread clearly shows that stitching programs do an excellent job... and these low-res (versions of your) pictures prove your point about perspective, but they do not prove that a 100 Mpx rectilinear pan-and-stitch (e.g.Nikon) DSLR image is as sharp as a 100 Mpx shift-and-stitch image created with a MFDB like a P65+ or a H3D22-60 with a view camera.

Any process that re-sampls an image, projects it, or passes an image through another piece of glass is bound to decrease resolution... but this effect might be imperceptible.

I appreciate that a cylinder pano pan-and-stitch image often looks better than anything you could achieve with a single shot camera (ignoring the Seitz roundshot), and I expect to use my MFDSLR or view camera for pan-and-stitch where appropriate.

I hope that in a few days I will have live view working on my camera, and I will be able to produce some test images.

One effect of pan-and-stitch or computer perspective (converging verticals) correction e.g. in tall buildings is that the resultant image is truncated in the direction of the converging lines... of course it might be possible for the stitching software to compensate for this, or you could manually stretch (PS distort) the image, but this too would tend to slightly degrade the image.

It you want to think of viewpoint perspective, think of a Kamara Obskura (there is one just up the road at Compton Verney) or a pin hole camera, and eliminate any thoughts about lens distortion.

View-point perspective...

In a (1980) Calendar shot of a tractor mounted mower I wanted the hill in the background to look bigger in relation to the tractor... and I was fortunate that I was able to use the telephoto instead of the standard lens: this evolved having the camera within two feet of the ground to get the hill in the right place relative to the tractor, but it produced the desired result.

Now... if I do not get out and mow the lawn my wife will not be happy.

[filmcapture]

This is actually an interesting argument, which reminds me of some basic theories of Optics I learned in high school and undergraduate (BTW, my background is engineering and physics). It's true some great photographers might not be proficient in such theories, and it does not mean photographers with handful theories are excellent either. Most of us learn these stuff from practice. Yes, seeing is believing, right? Happy shooting!

[pcunite]

This is a pretty good write-up (with pictures) on the subject.

http://photography-on-the.net/forum/showth...dof+perspective

[elf]

To me, it's simple.  Set the tripod and camera to the position that gives the desired perspective/composition.  Choose the lens that will give the desired detail (resolution), which also dictates how many frames will be needed. If you don't have a lens that will cover the FOV in one image you will need to stitch.  If more detail is desired than a short focal length lens will produce, use a longer focal length.

For in-camera composition fanatics, set the composition with a wide angle lens, then use a longer focal length lens to gain the detail (resolution).

[Panopeeper]

they do not prove that a 100 Mpx rectilinear pan-and-stitch (e.g.Nikon) DSLR image is as sharp as a 100 Mpx shift-and-stitch image created with a MFDB like a P65+ or a H3D22-60 with a view camera

The stitched result from DSLR shots can be less or more sharp than the MFDB shots. It depends not only on the DSLR camera (and lens quality), but on the selected focal length. One can choose a lens, which gives a much higher overall resolution than the MFDB shot; the interpolation of the warping process will be better, if shrinking instead of stretching is involved. In other words, a 100 Mpix pano can be created from 200 Mpix sources plus overlaps.

Note, that the source frames of the stitching should never be sharpened, except for scene referred sharpening if starting out with raw images; only the finished pano should be sharpened.

[Dick Roadnight]

The stitched result from DSLR shots can be less or more sharp than the MFDB shots. It depends not only on the DSLR camera (and lens quality), but on the selected focal length. One can choose a lens, which gives a much higher overall resolution than the MFDB shot; the interpolation of the warping process will be better, if shrinking instead of stretching is involved. In other words, a 100 Mpix pano can be created from 200 Mpix sources plus overlaps.

Note, that the source frames of the stitching should never be sharpened, except for scene referred sharpening if starting out with raw images; only the finished pano should be sharpened.

If you do a rectilinear shift-and-stitch with a digital back and a good lens, and print direct from the original file at your printer's best dpi there is no scaling, stretching or warping, (both the H3D11-60 and the P65+ will print 24" @ 360 original camera pixels per print inch). You might need to pan-and-stitch 400 Mpx   to look as good as a 100 Mpx MF shift-and-stitch.

[Panopeeper]

You might need to pan-and-stitch 400 Mpx   to look as good as a 100 Mpx MF shift-and-stitch.

I'm afraid this is speculation without any factual basis.

[BernardLanguillier]

Bernard...

I am not trying to pick an argument with you or anyone.

I agree that your post above in this thread clearly shows that stitching programs do an excellent job... and these low-res (versions of your) pictures prove your point about perspective, but they do not prove that a 100 Mpx rectilinear pan-and-stitch (e.g.Nikon) DSLR image is as sharp as a 100 Mpx shift-and-stitch image created with a MFDB like a P65+ or a H3D22-60 with a view camera.

Hi Dick,

This thread is not about quality, it is about the ability of stitching softwares to deliver results that are geometrically similar to those of a single lens. It is also about confirming with facts that theory that perspetive is only affected by camera/subject and that wide angle lenses do not bring any magic quality to the table.

As far as quality goes, there are links to full size versions of the images in my post as well. I keep thinking that we are very close to backs at pixel level, but we don't need to be. I could have shot this with a 100mm lens, it would have taken 3 times the time and the resolution would have been 3 times higher.

Cheers,
Bernard

[BernardLanguillier]

You might need to pan-and-stitch 400 Mpx   to look as good as a 100 Mpx MF shift-and-stitch.

I agree that you will indeed need more pixels from a well executed D3x stitch to equal the sharpness of a back, but 4 times more?... I would be interested in hearing where you got this figure from.

Assuming that everything was done perfectly on both sides, my view is that the ratio is probably never higher than 1.5 even when comparing files on screen, and even lower when comparing prints. I was pretty amazed a few months ago to see that a 70cm wide pano printed awesomly well although the individual d2x images making it up had suffered from significant blur because of a tripod mount issue.

Considering that focusing a back very accurately in the field is not an easy task, there are in fact probably many real world cases where the ratio is close to 1 or even smaller than 1.

Anyway, I'll be spending the required 800 US$ in the coming weeks to rent a P65+ with a recent Mamiya lens and do my own comparison. I will be honnest about what I see.

Cheers,
Bernard

[Dick Roadnight]

You might need to pan-and-stitch 400 Mpx to look as good as a 100 Mpx MF shift-and-stitch.

I agree that you will indeed need more pixels from a well executed D3x stitch to equal the sharpness of a back, but 4 times more?... I would be interested in hearing where you got this figure from.

This is speculation without any factual basis, and this is why I used the word "might"... but, as always, it depends if you are comparing apples with apples... or comparing a picture produced from several pan-and-stitch stretched images taken with a camera with and anti-aliasing filter with one from two shift-and-stich images taken with an AA free camera. To evaluate the pan-and-stitch process, we would need to use the same quality of camera for both pictures.

I think that 4* might be correct for comparing a MFDB/apo digitar 210mm image with a stitched image from a 480mm large format sheet film lens, even if it was shift-and-stitched. I hope to be acquiring an apo-digitar 210 sometime soon, and I have a 480.

Assuming that everything was done perfectly on both sides, my view is that the ratio is probably never higher than 1.5 even when comparing files on screen, and even lower when comparing prints.
Considering that focusing a back very accurately in the field is not an easy task, there are in fact probably many real world cases where the ratio is close to 1 or even smaller than 1.

The focusing on the H3D is OK, and I do not expect to do much view camera work without live view.

Anyway, I'll be spending the required 800 US$ in the coming weeks to rent a P65+ with a recent Mamiya lens and do my own comparison. I will be honnest about what I see.

Cheers,
Bernard

We look forward to reading you comments... if you were not on the other side of the world, we could team up to do some comparisons, and save you the $800!

I have started a thread to compare the P65+ to the H3D11-60.

[cmi]

I dont know why you guys act so complicated because this side-issue is trivial. Bernard is of course right. Oversampling (downrezzing) solves everything. It's obvious that you can get every quality from it, even exceeding MFDB, no doubt.

From my experience with pixels I can tell I need maybe a 50-75% downres along with some sharpening gets me really high qualtity samples. If I want to achieve the ultimate pixel quality I ofc want to downsample as much as possible, through there is a point where you dont get much benefit beyond. That depends on the source image, but generally I'd say, 65% gets you pretty close to perfect. I dont own a MFDB, but this is common knowledge if you push your pixels and are used to look at the pixel level.

I mean thats exactly one advantage of stitching, you just massively oversample at shooting time. So you can get from a much much lower camera exactly the same or much better resolution than from a much more expensive cam. Trivial. Oversampling solves everything, so its just the question how much data you want/can/afford to practically capture regarding processing and storage. And for inferior source pixel sharpness of smaller cams, from a certain downscale factor on you just cant tell.

The disadvantage of achieving such high quality from smaller sensors is of course that you have to shoot very big and to downsample quite a bit. Depends on circumstances and camera.) But if you do this, you indeed end up with very very high quality files equal to or exceeding mfdb. I can say this without owning a mfdb because the quality I get when I downsample image data gets equal to my renderings. And rendered files, this is the cleanest quality you will **ever** get.

Christian


//Edit: And I could even understand if someone where a bit frustrated about it. If I own an expensive system and then realize I could pretty much workaround with lower tech stuff to get the same q results in some/many instances (or maybe in nearly all instances given my work)... And if I (maybe) only need the big MFDB file seldom... And if I had to struggle to get my high end system... then the issue could indeed be frustrating for me!

And I dont argue against mfdb - that would be CHILDISH. Its just and only about the quality issue in this particular regard in this particular discussion.

[Guillermo Luijk]

I think to understand the question under discussion, it's is mandatory and sufficient to understand about image projections in pano stitching software. A great place to learn about these topics: PT Assembler projections.

With PT Assembler one can even transform one single image from its rectilinear original state (straight from the camera+rectilinear lens), into any other kind of projection. This is sometimes very interesting for example in strongly distorted WA shots to make them more pleasant to the eye. Moreover you learn a lot in the process.

Regards.

PS: I agree Bernard's sensor has some dust in it.

[BernardLanguillier]

PS: I agree Bernard's sensor has some dust in it.

At least one person is following!  

Cheers,
Bernard

[bradleygibson]

Jeff,

Thank you for the comprehensive reply.  I think you have summed up the answer to Nick's question very nicely.

-Brad

First it's probably a good idea to agree on some terminology. What I would call geometric distortion has nothing to do with perspective or even focal length, except to the extent that certain focal lengths tend to have certain types of geometric distortion due to lens design contraints. Namely, telephotos often have pincushion, while wide-angle lenses are more likely to have barrel distortion. These distortions are optical flaws in lenses though, and have nothing to do with perspective. In fact two 20mm lenses can have varying amounts of barrel distortion.

Geometric distortion is not what causes elements at the edge of a wide-angle shot to be stretched out. That's rectilinear distortion, or perspective distortion (I've seen it called both). Rectilinear distortion has everything to do with perspective and field of view. Note I said field of view, not focal length. Whether you get that perspective and field of view from a single wide-angle shot or by shooting several shots with a longer lens and then stitching them using a rectilinear projection, the resulting perspective will be the same.

It's true that if you just take several shots with a longer lens and align them you won't get the same look as a wide-angle shot. But that's not what pano software does; it projects the images, and as part of that process the outer portions of the resulting image will get stretched just like they do in a wide-angle shot. The wider the field of view, the more the edges are stretched; which is why even though you can shoot a much wider FOV with stitching than with the widest of lenses, you probably won't want to use a rectilinear projection for extreme FOV's.  Note that other projections such as cylindrical or spherical do not produce the same results as a rectilinear projection.

You don't need panos/stitching to prove that perspective is determined solely by position and not focal length. Take a zoom lens such as the 17-40 and mount it on a tripod. Take a shot at 17mm, and another at 40mm. On the computer, crop the 17mm image so that it covers the same part of the scene as the 40mm image, and you'll find that the perspective of the two resulting images (eg the spatial relationships and relative sizes of objects in the scene) will be exactly the same. I think DPReview even has an article covering this with sample shots if you don't feel like doing the experiment yourself. It's incontrovertible fact, and it really does amaze me how many photographers refuse to grasp this concept. Usually it's 35mm photographers who had never shot any other formats before digital came along, and you'll hear them saying something like  "I don't like DX format because I can't use a 60mm lens to get the same pleasing portrait perspective that my 85mm lens gave me on 35mm". Try shooting that 85mm portrait on 4x5" film and then tell me that it's the focal length that determines the perspective....

[Ray]

You don't need panos/stitching to prove that perspective is determined solely by position and not focal length. Take a zoom lens such as the 17-40 and mount it on a tripod. Take a shot at 17mm, and another at 40mm. On the computer, crop the 17mm image so that it covers the same part of the scene as the 40mm image, and you'll find that the perspective of the two resulting images (eg the spatial relationships and relative sizes of objects in the scene) will be exactly the same. I think DPReview even has an article covering this with sample shots if you don't feel like doing the experiment yourself. It's incontrovertible fact, and it really does amaze me how many photographers refuse to grasp this concept. Usually it's 35mm photographers who had never shot any other formats before digital came along, and you'll hear them saying something like  "I don't like DX format because I can't use a 60mm lens to get the same pleasing portrait perspective that my 85mm lens gave me on 35mm". Try shooting that 85mm portrait on 4x5" film and then tell me that it's the focal length that determines the perspective....

All this is true only if one ignores the definition of a wide-angle lens. In your example above where you write "crop the 17mm image so that it covers the same part of the scene as the 40mm image" you are in effect saying, 'turn the 17mm lens into an equivalent 40mm lens and it will behave like a 40mm lens'. This line of reasoning could be considered a tautology. The fact is a 40mm lens is not a wide angle lens on the 35mm format, neither on full frame nor on APS-C format. Any 17mm shot on a 35mm camera, of a scene at a specified distance will give one a different perspective of the scene than a 40mm shot of the same scene from the same distance, simply by virtue of the fact it is a different image with a wider field of view.

For example, in the 40mm shot there might be no visible background, no clues as to the closeness or relative size of the subject because all distant objects in the scene were not captured. I might be photographing, for example, a bas-relief on a small section of wall in a ruined temple. Using a 40mm lens, I might capture only the bas-relief and a bit of surrounding stone-work. In the 17mm shot, with lens pointing in the same direction, there will be vastly more 'picture information' surrounding the subject on all sides. There might be open fields in the background with small figures wandering around which will produce the effect that the bas-relief is huge and/or that one is very close to it. In this sense, the perspective of a 17mm shot is not the same as that of a 40mm lens at the same distance, on the same format of camera, as the eye sees it.

Whilst a 40mm lens is not a wide angle lens on the 35mm format, it is on full frame Medium Format. When we stitch images, we are in effect, increasing the format size of our camera, just as when we crop an image we are in effect increasing the focal length of our lens.

In my view, perspective is determined by both position and the focal length of the lens in relation to the camera format size.

[Panopeeper]

Any 17mm shot on a 35mm camera, of a scene at a specified distance will give one a different perspective of the scene than a 40mm shot of the same scene from the same distance, simply by virtue of the fact it is a different image with a wider field of view.

It must be a great fun to join to a discussion with a different understanding of the basic terminology than most others have.

[cmi]

Excuse me, can anybody of you explain in simple words, possibly in one sentence, what the discussion is about? I see only random facts. The goal of the discussion is unclear to me. Sure its somehow about dof perspective and stitching, but the point of all of this got somehow out of sight.

[BernardLanguillier]

Excuse me, can anybody of you explain in simple words, possibly in one sentence, what the discussion is about? I see only random facts. The goal of the discussion is unclear to me. Sure its somehow about dof perspective and stitching, but the point of all of this got somehow out of sight.

I believe that the true underlying topic of this thread is the relationship between pink shrimps penis lenght and climate change in sub-tropical Russia. But I have been wrong before.

Cheers,
Bernard

[Ray]

It must be a great fun to join to a discussion with a different understanding of the basic terminology than most others have.

My first digital camera was the cropped format D60 and I learned very quickly that a 50mm lens on that camera would produce not only the same perspective as an 80mm lens on full frame 35mm, from the same distance, but also the same FoV.

But I'm a bit uncomfortable with the notion that any lens from the same position, from 12mm to 600mm (if one were within the minimum focussing distance of the 600mm lens) would produce an image with the same perspective. They would be different images. If the 600mm shot included only a left nostril, then the 12mm shot might include the whole person in a large room.

If one were to compare A3 size prints of both images, it would seem a bit odd to me declare that the perspective of both images is the same, although I understand if the 12mm shot were sufficiently high resolution, and the 600mm shot were very low resolution, I could crop the 12mm shot and get a left nostril which looked very similar to the 600mm shot.

[BernardLanguillier]

But I'm a bit uncomfortable with the notion that any lens from the same position, from 12mm to 600mm (if one were within the minimum focussing distance of the 600mm lens) would produce an image with the same perspective. They would be different images.

They would be different crops but feature the same perspective, unless we all agree that the definition of the word perspective needs to be changed it might be wise to stick to its currently accepted definition.

Think of a world where some people use the word "pink shrimp" as meaning "great white shark" while others would use "great white shark" when they mean "corruption in sub-tropical Russia". Wouldn't that be confusing?

Cheers,
Bernard