Perspective and Distortion

[marc aurel]

This geometrically challenged photo unit (me) is slowly coming to grips with how a camera system functions. One interesting side effect of that is a greater appreciation of architectural painting. Seems to me that painters have freedom to introduce localised distortions that make an image perceptually "better" even as they violate the model of what a projection should ook like. @marc aurel 's church example is a photographic illustration of how that might work - but doesn't. The camera can't paint a bit of barrel distortion in at just the one place it looks better. In contrast, brush artists (the really good ones) can do barrel here and keystone there and .... you get the idea. Look at this image from Cooper. All sorts of subtle geometrical tweaking going on in there and the result is pleasing. I need to try to find a photo of the same building to compare.

As @Shashin points out, there is a long history of using optical distortion in artistic photography but, I suppose, those experiments are not particularly relevant to architecture. There, the best we can do as photographers is to apply movements and we may have to learn to like the results because of the constraints of the optical model. At least that is what this novice comes away with.

Don't say I can't cheat like a painter ;-)
Although I have to say I hate it and have done that only with 2 or 3 images in my whole career. Usually I go back as far as I can. But when I hit the wall in my back and there are distorted obejcts at the edge of the frame – then there is photoshop. An example:

Original image:


I just did not like the shape of the lamps at the sides. So I transformed those in photoshop and collaged them into the original image:


Crop of the lamp on the right in direct comparison:


I just transformed the lamp a little bit. And I think it worked for that image, because there is not much reference in the rest of the image where you can see that there is something wrong. But this is an emergency workaround, and it's cheating.

 

[MGrayson]

Don't say I can't cheat like a painter ;-)
Although I have to say I hate it and have done that only with 2 or 3 images in my whole career. Usually I go back as far as I can. But when I hit the wall in my back and there are distorted obejcts at the edge of the frame – then there is photoshop. An example:

Original image:
View attachment 209618

I just did not like the shape of the lamps at the sides. So I transformed those in photoshop and collaged them into the original image:
View attachment 209619

Crop of the lamp on the right in direct comparison:
View attachment 209620

I just transformed the lamp a little bit. And I think it worked for that image, because there is not much reference in the rest of the image where you can see that there is something wrong. But this is an emergency workaround, and it's cheating.

There is no cheating in art! Very nicely done.

 

[Ben730]

Don't say I can't cheat like a painter ;-)
Although I have to say I hate it and have done that only with 2 or 3 images in my whole career. Usually I go back as far as I can. But when I hit the wall in my back and there are distorted obejcts at the edge of the frame – then there is photoshop.

I do that often when I shoot bathrooms. In most cases, there is so little space that only super-wide angles work, which means that the bathroom furniture is often warped in the ugliest possible way.

 

[Ben730]

@pegelli + @dchew

In the thread "Fuji GF 30mm Tilt Shift and Rodenstock 32mm HR Digaron-W Lens Comparison" I made a test.
1. 14 mm with Fuji GFX100s cropped to ≈ APS-C ≈ 24.5 mm x 18.4 mm sensor size
4. 55 mm stitched of 9 pictures with IQ 3 100 ≈ 93.4 mm x 70 mm sensor size
So was the difference not due to the different lens focal lengths, but to the imperfect alignment (due to different cameras) of the two shots?
I find number one more squeezed than 4.

 

[MGrayson]

@pegelli + @dchew

In the thread "Fuji GF 30mm Tilt Shift and Rodenstock 32mm HR Digaron-W Lens Comparison" I made a test.
1. 14 mm with Fuji GFX100s cropped to ≈ APS-C ≈ 24.5 mm x 18.4 mm sensor size
4. 55 mm stitched of 9 pictures with IQ 3 100 ≈ 93.4 mm x 70 mm sensor size
So was the difference not due to the different lens focal lengths, but to the imperfect alignment (due to different cameras) of the two shots?
I find number one more squeezed than 4.
View attachment 209621View attachment 209622

You can see from the way foreground objects (broncolor A4's) overlap the ones behind them that the second image is taken from further away, and that changes the image slightly, but visibly. In the bottom image, the boxes on the far right are cropped a bit out of the frame. In the top image, they crowd in a bit.

 

[dchew]

So, this means if you shoot GFX + 30TS you have to go only 10mm forward 8with the camera/sensor to shoot exactly the same with IQ3 100 + 40HR?

Although @pegelli and @marc aurel answered very well, I will pile on

Yes. A good example for comparison would be the sk35 w/ 100c vs the sk43 w/IQ3100. If you place the lenses for those two systems in exactly the same place, they will give you precisely the same image. The IQ3100 sensor will be about 8mm behind the HB 100c but the lenses will be in the same physical location to give you the same view.

Dave

 

[pegelli]

@pegelli + @dchew

In the thread "Fuji GF 30mm Tilt Shift and Rodenstock 32mm HR Digaron-W Lens Comparison" I made a test.
1. 14 mm with Fuji GFX100s cropped to ≈ APS-C ≈ 24.5 mm x 18.4 mm sensor size
4. 55 mm stitched of 9 pictures with IQ 3 100 ≈ 93.4 mm x 70 mm sensor size
So was the difference not due to the different lens focal lengths, but to the imperfect alignment (due to different cameras) of the two shots?
I find number one more squeezed than 4.

As Matt said, a slightly different central axis as well as a slightly different distance make this comparison too inaccurate to draw any conclusion about wide angle corner distortion with a rectilinear projection. For me the first difference I saw is the eye piece of the camera near the top edge on the right. In the first photo it's touching or even shaded by the metal column, in the second there is a significant space between the two. Also the Broncolor grafit A4 on the left shades the chair leg at the edge differently, and there are many more small differences that can be spotted to conlude you need to repeat this test with much more accurate placement of the lens entry pupil of both versions. I also wouldn't bring in the complexity of stitching an image, software programs might stretch/squeeze some pixels to make all the individual frames fit perfectly and who knows what it does to the exact resulting rectiliniar projection.

 

[pegelli]

@Ben730 , another question I would have is how you shifted for your second image. You can keep the camera in one place and shift the lens left/right/up/down or keep the lens position constant and move the camera/sensor behind it to get your 9 shots. I think the latter method is the correct one to do the comparison, the first method will (I think) give less WA distortion of the rectiliniar projection since you're moving the entry pupil in the direction of those corners.

 

[corvus]

The 2014 MF Image thread was branching off into the issues of the perception of distortion that comes (largely) from wide-angle and shifted lenses. So here's a place to discuss it.

I'll prepare a short exposition on the subject, but right now, I'll just leave the guiding principle: Every photo looks normal when viewed from the camera's location. A wide angle photo had a camera close to the building. Place your eye close to the image, and it will look normal. A shifted image has the camera located closer to one of the edges of the image. Put your eye there at the appropriate distance and you'll see the image "normally".

Our perception of distortion comes largely from viewing the final image from the "wrong" place.

A different way to see it. Stand right behind a small light source. Any object's shadow will look to *you* just like its own shape - like a thin dark halo. The actual shadow may be very elongated on the ground, but it won't look like that to you. Go away from the light source, and the shadows will start to look distorted. This is exactly the same principle.

Pictures and verbiage to follow...

Matt

I would like to establish as first takeaway that our retinas are curved by Mother Nature (meaning perpendicular incident angles across our FoV) so we will perceive unnaturally distorted projections of equidistant geometric forms onto to edges of a film plane via steep incident angles as disturbing, especially if it is not a central perspective because we don't, as humans, have that problem ... so that influences our subconscious interpretation of unnatural looking, overly distorted known shapes.

Since curved sensors do not exist yet, the easiest way to get more natural spacing of geometry is to use longer focal lengths and move back from the subject until you cover it fully.

Another key aspect is the difference between central and two point perspectives and placing of the focal point in the scene during composition.

If you can avoid convergence towards the upper or bottom parts of the image and especially in interior scenarios can work with central perspective – that's IMHO harmonious. In practice, you cannot always bring a ladder to a scene, but you can still seek out ideal placement and have that in mind.

So composition can save an overly wide lens that may be required in an interior scene (e.g. room with 23 HR).

If you don't employ a central perspective placing of the focal point and subjects along harmonious points within a composition helps a lot to create pleasing imagery, ie adherence to golden cut, thirds, etc.

A very good thread that exactly supports my previous experiences and considerations, Matt! ... & good contributions from Paul, shashin, marc ... too!
We architects attach great importance to the fact that spatial structures are depicted approximately as we perceive them locally. This corresponds to an image angle of between 40°-55°, which is close to the natural image or viewing angle. For a focal length that is as balanced as possible, I use the rule of thumb that it should roughly correspond to the diagonal of the image or selected format. In full-frame format, for example, this would be 40-50mm - which is also the reason for the commonly used term "normal focal length". Then the proportions and the depth of space appear approximately as in the real perception of space. We then also perceive a square as a square and not as a rectangle A little wider is still okay, but I would not prefer to get as much as possible in the picture with extremely large picture angles if the picture proportions suffer as a result. Then I'd better concentrate on a detail ...

Torsten

 

[Ben730]

@Ben730 , another question I would have is how you shifted for your second image. You can keep the camera in one place and shift the lens left/right/up/down or keep the lens position constant and move the camera/sensor behind it to get your 9 shots. I think the latter method is the correct one to do the comparison, the first method will (I think) give less WA distortion of the rectiliniar projection since you're moving the entry pupil in the direction of those corners.

I moved the back with Cambo WRS. + - 20 mm left/right, + - 15 mm up/down

A good example for comparison would be the sk35 w/ 100c vs the sk43 w/IQ3100. If you place the lenses for those two systems in exactly the same place, they will give you precisely the same image. The IQ3100 sensor will be about 8mm behind the HB 100c but the lenses will be in the same physical location to give you the same view.
Dave

I don't have a Hassy back and I sold my 35XL and the 43 XL long time ago, but I have an IQ1 50 and an IQ3 100 with 23HR, 32HR, 47XL, 55digital, 72XL, a GFX100s with most lenses from 20-120, Nikon adapter from 14-200 mm incl. the 19 mm TS etc.
Let me know if you wish a comparison. I think a comparison only makes sense within the same camera system,
as the smallest deviations in the camera/sensor position lead to similar misjudgements that I already made in the first test.

 

[pegelli]

This church has eaten too much and has gone fat ;-)

Thanks for this example Marc, very illustrative. Indeed in the "perfect" rectiliniar projection the statue becomes stretched and the reason for that is clearly explained by Matt's diagrams. I was wondering if there is a software that can progressively squeeze pixels in in one direction (in this case vertical, most at the top, none at the bottom of the photo) and not modify them in the horizontal direction (which is what your barrel distortion trick does. That way the statue will look less squeezed, the church doesn't become "fat" and the normal view of the lower area (width x height) is maintained. I think such a method could be beneficial for both heavily shifted photos as well as software keystone corrected images. If we would apply that to the third picture in Matt's post #15 (squeeze in one direction from from both sides and not in the middle) all spheres would again become round, allthough they would still overlap with the sidewalk and the road would become narrower.

 

[dchew]

I don't have a Hassy back and I sold my 35XL and the 43 XL long time ago, but I have an IQ1 50 and an IQ3 100 with 23HR, 32HR, 47XL, 55digital, 72XL, a GFX100s with most lenses from 20-120, Nikon adapter from 14-200 mm incl. the 19 mm TS etc.
Let me know if you wish a comparison. I think a comparison only makes sense within the same camera system,
as the smallest deviations in the camera/sensor position lead to similar misjudgements that I already made in the first test.

Unfortunately, none of those combinations match up too well:

 

[Shashin]

Finding a perfect projection in photography is like Home Depot: it will almost have what you want. It usually comes down to where you want to make the compromise(s).

However, it is not always a camera-side problem. The print and its presentation can be used to minimize some of this. Angles that seem extreme when small, soften when big.

 

[tenmangu81]

[....]
Although I have to say I hate it and have done that only with 2 or 3 images in my whole career. Usually I go back as far as I can. But when I hit the wall in my back and there are distorted obejcts at the edge of the frame – then there is photoshop. An example:

Original image:
View attachment 209618

I just did not like the shape of the lamps at the sides. So I transformed those in photoshop and collaged them into the original image:
View attachment 209619

Crop of the lamp on the right in direct comparison:
View attachment 209620

I just transformed the lamp a little bit. And I think it worked for that image, because there is not much reference in the rest of the image where you can see that there is something wrong. But this is an emergency workaround, and it's cheating.

I really love this image, Marc. But the shape of the lamps, even when transformed with Photoshop, hurts me a little bit. I am sure if I were at the position of your camera when you took the picture, I would see these lamps as spheric. But your example illustrates perfectly the difference between the projection camera does and what our eyes see (an interpret).

 

[marc aurel]

I really love this image, Marc. But the shape of the lamps, even when transformed with Photoshop, hurts me a little bit. I am sure if I were at the position of your camera when you took the picture, I would see these lamps as spheric. But your example illustrates perfectly the difference between the projection camera does and what our eyes see (an interpret).

Actually the lamps are not spheric, they are eliptical in cross-section (less height than width).
But nevertheless I see your hurt as a valid reaction, because I suspect it's more because of the "diagonal" appearance of the distortion. Sometimes when you want to take a certain image, you have to live with a little hurt.

 

[marc aurel]

There are cases where you just have to live with distortion because there is no other way to get a certain shot.

This building is the "Umlauftank 2" in Berlin by the architect Ludwig Leo. It is a research building that was built between 1969 and 1975. The large tube is coated in pink polyurethan foam, inside they accelerate water to 10m/s that goes round in circles through the tube. The blue box is 5 floors high and at it's bottom there is a basin where you can put in ship models in the accelerated water to test their shapes for flow resistance. I took photos after the renovation. The machinery is still functional. Look at the scale of the lorry in front of it to see how large this architecture is. The tube has an overall length of 120m and contains 3.300t of water.

If you look at the first shot – yes of course, the tube segments are heavily distorted, they have a circular section in reality. And the shot was very wide with a lot of shift. To reduce distortion I got as far back as possible, but behind my position there were dense trees. And I tried to get a bit higher, so I stood on a houseboat that lies in the canal next to the building. And I used a large tripod.

A drone might have helped to get even higher. But I had the intention not to get away too far from a pedestrians perspective.
So I live with the distortion. Because this shot is basically the one I wanted.

On the second image you can see how this building is situated at the edge of "Tiergarten", the large central park in Berlin. I took it from the roof of a university building nearby. A loooong focal length with no distortion ;-)

 

[tenmangu81]

Thanks Marc for this information and images. I intend to visit Berlin by April and I'll try to have a look at this fantastic research building. And shoot it !!

 

[rdeloe]

I think the theme of this thread is, "In photography, perfection is the enemy of done."

 

[MGrayson]

I think the theme of this thread is, "In photography, perfection is the enemy of done."

In everything, really. I was fond of telling new hires "It's suboptimal to optimize."
(For the less imaginative, "If the overnight runs don't finish overnight, that would be bad.")

 

[dchew]

On the second image you can see how this building is situated at the edge of "Tiergarten", the large central park in Berlin. I took it from the roof of a university building nearby. A loooong focal length with no distortion ;-)

Marc, that second photo really points out how wide the first image is. I think I see the house boat on the other side of the river; amazing!

In your statement above, I would argue, "... a looooong distance from the subject with no distortion."


Dave