Stitched Nikon to emulate medium/large format capture?

[BernardLanguillier]

Stitching workflow seems to have more in common with shooting8x10.

Precisely so. Just like scanning backs, this isn't for everybody, but if you are demanding the best possible image quality stitching is the only option (whether you stitch with a DSLR or a back by the way).

Cheers,
Bernard

[hjulenissen]

How do you stitch several images from a single camera to most closely emulate what you could have gotten (geometrically) with a single, larger sensor/lense? tilt/shift lense and moving the camera along the image "projection plane"?

-h

[Bart_van_der_Wolf]

How do you stitch several images from a single camera to most closely emulate what you could have gotten (geometrically) with a single, larger sensor/lense? tilt/shift lense and moving the camera along the image "projection plane"?

Yes, stitching shifted tiles provides the closest emulation of a single larger sensor plane, and gives identical results in an untilted focus plane. For geometrical emulation, one can also  use images with adequate depth of field that were rotated through the entrance pupil of the lens (one just needs to use a different projection model to recreate a flat plane).

The latter solution requires resampling of the image, which might sacrifice some resolution (unless compensated for with a longer focal length to oversample the scene, and a downsampling for same size/resolution output). A benefit of the rotation method is that one can use the higher quality part of the image circle to stitch, and it's generally faster in execution at capture time (slower in postprocessing). It also allows to use very wide (>110 degrees) angles of view.

Cheers,
Bart

[hjulenissen]

So all of these could give (principally, at least, given flawless lenses, sensors, etc) geometrically equal results, given appropriate stitching (and perhaps constraints on scene distance?)
A) Some large format sensor/lense
B) A smaller format sensor/lense tilt/shifted appropriately
C) A smaller format sensor/lense rotated "inappropriately" about the sensor centre point

Did anyone do the comparision and post example pictures? =)

-h

[elliot_n]

So all of these could give (principally, at least, given flawless lenses, sensors, etc) geometrically equal results, given appropriate stitching (and perhaps constraints on scene distance?)
A) Some large format sensor/lense
B) A smaller format sensor/lense tilt/shifted appropriately
C) A smaller format sensor/lense rotated "inappropriately" about the sensor centre point

Did anyone do the comparision and post example pictures? =)

-h

I started this thread, and I now have a bit more experience with stitching Nikon D700 images (in a 3x3 grid) to emulate a single large format film capture.

Regarding geometry, the only difference is that the stitched image exhibits zero lens distortion (barrel/pincushion). Geometrically, it is perfect. (A 5"x4" film capture, or a single image from a medium format digital back, will always exhibit a certain amount of lens distortion.)

Likewise, the stitched image exhibits no vignetting and no corner softness.

In theory this is all good news, but in practice stitched images can look rather clinical.

I've recently been experimenting with adding a little barrel distortion, vignetting and grain to the final stitched image; knocking it back a little, to make it look more photographic.

[MalcolmL]

Good - I agree with your comments.
I did an interesting experiment with stitching a maritime dusk shot using an old pentax 55mm f1.8 smc lens on a Sony a850 FF camera. I took 3 in a row in portrait orientation and then printed up on a Canon Pixma pro printer at max resolution on A3 + size paper. That is 329 x 483 mm size.

The results were razor sharp. I then cropped out the central 50% of the stiched image and printed at the same size. Still razor sharp and detail was rendered that was not there in the uncropped print.

Message - you do not need massive stitches unless you are doing massive prints on a huge printer. If you try to do say 8 or 9 or 12 stich prints on an A3+  there will be detail you will never see.
I only do 3 in a row stiches now and it gives me more than enough.
Kind regards
Malcolm

[OldRoy]

Elliot_n said:
I've recently been experimenting with adding a little barrel distortion, vignetting and grain to the final stitched image; knocking it back a little, to make it look more photographic.

Please reassure us that this is a joke.

Roy

[elliot_n]

No joke (though I was hoping to get a rise from someone). As I said, an experiment. I'm talking about quite a wide angle field of view; equivalent to a 24mm lens or wider (in 35mm terms). Such a view can look especially stretched when rendered with zero lens distortion. Adding a little barrel distortion has the effect of pushing back the corners a little. I guess it's what I'm used to, shooting nikon primes from 20mm to 50mm - they all exhibit a little barrel distortion.

[hjulenissen]

...Regarding geometry, the only difference is that the stitched image exhibits zero lens distortion (barrel/pincushion). Geometrically, it is perfect. (A 5"x4" film capture, or a single image from a medium format digital back, will always exhibit a certain amount of lens distortion.)
...

By "geometry" I was actually thinking about the physical effects that must be different (even using hypothetical perfect optics) when:
A)Shooting some object at a distance r1 using a "small" sensor/lense at focal length f1.
B)Shooting that same object at distance r1 using a "large" sensor/lense at focal length f2 to get the same framing.
C)Shooting that same object at some other distance, r2 using a "large" sensor/lense at focal length f1.

I was under the impression that B) and C) was sufficiently different from A) that people were willing to pay a lot (in money, weight, inconvenience) just to get that difference, even ignoring noise/dynamic range/... differences? If that really is so, I wonder if stiching can emulate that "magic factor"?

-h

[routlaw]

Indeed, this is one of the major advantage of cylindrical stitching/planar projection compared to T/S stitching. You always use the central portion of the lens and the corners of the resulting pano are just as sharp as the center.

Some very experienced Japanese landscape photographers who had never seen large panos before couldn't believe their eyes when they saw some large prints at a show I did a few months ago at the Nikon salon in Tokyo. They had a hard time dealing with the fact that corners were just as sharp as the center portion. The reproduction of the scene was so real that it had almost lost its photographic dimension to them. It did trigger some interesting discussions though.

I have tried both extensively and have basically given up on shift pano, it is slower, less scalable and offers lower image quality.

I just discovered this thread which holds some interest to me, and am most curious about your comment above Bernard.

I understand the limitations of shift stitching though this is somewhat dependent on the lens and camera system. But to my way of thinking it would seem all of the stretching, contortion, shrinking and interpolation of pixels involved with cylindrical stitching would be far more problematic than simple shifting of a lens. In the past I have used both my 85 PC lens as well as the 24 PC with very good results though the 24 is more limited at the corners of an extreme shift/rise & fall. Likewise I have used my Nikons with the Cambo Ultima conversion kit and Rodenstock digital lenses to stitch up to 9-12 frames with great success and no image degradation at the corners what so ever. Granted this is a tedious experiment not to mention the huge weight issue of carrying gear like this into the field. Not everyone has a Ultima they would lug around in the first place, and in fact even I find it more trouble than its worth.

So you are saying absolutely even with the amount of interpolation of pixels with cylindrical stitching this is still a better method than using shifts?

[Bart_van_der_Wolf]

By "geometry" I was actually thinking about the physical effects that must be different (even using hypothetical perfect optics) when:
A)Shooting some object at a distance r1 using a "small" sensor/lense at focal length f1.
B)Shooting that same object at distance r1 using a "large" sensor/lense at focal length f2 to get the same framing.
C)Shooting that same object at some other distance, r2 using a "large" sensor/lense at focal length f1.

I was under the impression that B) and C) was sufficiently different from A) that people were willing to pay a lot (in money, weight, inconvenience) just to get that difference, even ignoring noise/dynamic range/... differences? If that really is so, I wonder if stiching can emulate that "magic factor"?

B can be emulated by A when A uses a longer focal length (f1>= f2) and stitching to make up for the wider field of view of B.
C has a different perspective because the entrance pupil moved, so everything changes, there is no direct comparison possible.

What is important for a technical quality comparison is, that the magnification factor is the same. That might reveal differences in the MTF of the lenses used. When the camera distance is varied then the magnification factor of foreground and background features changes, also relative to eachother. The magnification factor determines where on the system MTF curve the detail and contrast are placed.

Cheers,
Bart

[elliot_n]

By "geometry" I was actually thinking about the physical effects that must be different (even using hypothetical perfect optics) when:
A)Shooting some object at a distance r1 using a "small" sensor/lense at focal length f1.
B)Shooting that same object at distance r1 using a "large" sensor/lense at focal length f2 to get the same framing.
C)Shooting that same object at some other distance, r2 using a "large" sensor/lense at focal length f1.

I was under the impression that B) and C) was sufficiently different from A) that people were willing to pay a lot (in money, weight, inconvenience) just to get that difference, even ignoring noise/dynamic range/... differences? If that really is so, I wonder if stiching can emulate that "magic factor"?

-h

(B) and (C) are totally different. People pay for the difference between (A) and (B). Can (B) be emulated with stitching? Yes, I think it can. The geometry of the image is certainly the same. The "magic factor"? What's that? Maybe the way the depth of field falls off? I think you get the same depth of field effect with a stitched shot. You're certainly working with shallower depth of field when you're stitching.

[elliot_n]

Some very experienced Japanese landscape photographers who had never seen large panos before couldn't believe their eyes when they saw some large prints at a show I did a few months ago at the Nikon salon in Tokyo. They had a hard time dealing with the fact that corners were just as sharp as the center portion. The reproduction of the scene was so real that it had almost lost its photographic dimension to them. It did trigger some interesting discussions though.

Right - this is why I've been experimenting with subtly degrading the quality of the final stitched image.

[BernardLanguillier]

I just discovered this thread which holds some interest to me, and am most curious about your comment above Bernard.

I understand the limitations of shift stitching though this is somewhat dependent on the lens and camera system. But to my way of thinking it would seem all of the stretching, contortion, shrinking and interpolation of pixels involved with cylindrical stitching would be far more problematic than simple shifting of a lens. In the past I have used both my 85 PC lens as well as the 24 PC with very good results though the 24 is more limited at the corners of an extreme shift/rise & fall. Likewise I have used my Nikons with the Cambo Ultima conversion kit and Rodenstock digital lenses to stitch up to 9-12 frames with great success and no image degradation at the corners what so ever. Granted this is a tedious experiment not to mention the huge weight issue of carrying gear like this into the field. Not everyone has a Ultima they would lug around in the first place, and in fact even I find it more trouble than its worth.

So you are saying absolutely even with the amount of interpolation of pixels with cylindrical stitching this is still a better method than using shifts?

Yes, that is what I have seen.

It depends on the resolution though. My view is that it is much faster to rotate a lens that to shift it, so you end up being able to take more images with cylindrical than with shift to cover a single scene/crop, which means you will have more resolution if required. The effect of some interpolation on image quality isn't that large in the first place and more pixels will totally compensate for this small loss.

Besides, you will end up having loss of sharpness and light fall off in the corners even with the best T/S lenses (I still own the Nikkor 24 PCE which is excellent even if it was now overtaken a bit by the new Canon). My view is that even if interpolations were to cause issues with cylindrical panos (which again I don't find to be the case), it would be an order of magnitude less of an issue compared to soft corners.

Regards,
Bernard

[BernardLanguillier]

No joke (though I was hoping to get a rise from someone). As I said, an experiment. I'm talking about quite a wide angle field of view; equivalent to a 24mm lens or wider (in 35mm terms). Such a view can look especially stretched when rendered with zero lens distortion. Adding a little barrel distortion has the effect of pushing back the corners a little. I guess it's what I'm used to, shooting nikon primes from 20mm to 50mm - they all exhibit a little barrel distortion.

You might want to try PTgui. It has the ability to control the amount of lateral compression near the edges of images when using planar projections.

I only use it when dealing with images too wide for a lens to handle, but this might help you achieve the effect you are looking for.

Regards,
Bernard

[elliot_n]

You might want to try PTgui. It has the ability to control the amount of lateral compression near the edges of images when using planar projections.

I only use it when dealing with images too wide for a lens to handle, but this might help you achieve the effect you are looking for.

Regards,
Bernard

Hi Bernard. I am using PTGui.

How do you use it to control lateral compression?

[elliot_n]

But to my way of thinking it would seem all of the stretching, contortion, shrinking and interpolation of pixels involved with cylindrical stitching would be far more problematic than simple shifting of a lens.

I've just been comparing a stitched image (a 3x3 grid) with the 9 source images from which it is made (shot at 65mm, 25% overlap, to give an overall view representing a 24mm lens (in 35mm terms)).

The side effects of PTGui's complex interpolations are modest. The central area of the image has been downsampled slightly, and thus looks sharper than the source images, whereas the corners have been upsampled a bit, and look slightly softer than the originals.

[BernardLanguillier]

Hi Bernard. I am using PTGui.

How do you use it to control lateral compression?

Elliot,

In the window enabling you to tune the horizon, change projection mode,... (forgot the official name of that window), you have 2 collapsible sliders at the bottom left of the screen.

They only show when you are in flat projection mode.

Regards,
Bernard

[elliot_n]

Elliot,

In the window enabling you to tune the horizon, change projection mode,... (forgot the official name of that window), you have 2 collapsible sliders at the bottom left of the screen.

They only show when you are in flat projection mode.

Regards,
Bernard

Many thanks, Bernard.

I had forgotten those sliders existed. When I looked at them previously I assumed they were compressing the image universally. But you're right - they compress the image more at the edges than at the centre.

This will be very useful when rendering wide-angle views.

Elliot

[OldRoy]

I try to stay out of discussions where the standard of technical knowledge is above, if not entirely over my head, however a couple of comments.

I find the assertion that adding distortion, such as geometric distortion and vignetting as mentioned here, to an image, in order to get a more "photographic" representation, well, perverse. As far as I can recall every assessment of lenses I have ever encountered regards the absence of these attributes as a positive value; indeed most people seem prepared to pay a large premium to acquire lenses that don't exhibit these characteristics - although obviously there are also many other criteria in play. Of course adding vignetting in pp is sometimes an aesthetic decision. But adding barrel distortion? Seriously?

I was interested in the mention of PTGui's ability to compress the lateral edge distortion (volume anamorphosis?) in extreme wide angle shots. I hadn't spotted it. I've been looking for a way to achieve this for shots taken at the wide end of my Nikon 14-24. DxO is the only application I know of that incorporates the feature but I'm not prepared to buy it for this ability alone.

I'm not at a computer with PTGui right now, so has anyone used this feature successfully on single extreme wide-angle shots?
Roy