Lumix 18-40 Lens Corrections worse on than off??

[CharlesH]

Thanks all for the good discussion, and the distinction between lens distortion and perspective distortion.

I'd like to add some comments. First, about Lightroom and lens profile correction with Panasonic lenses, and with the 18-40mm lens. For a JPEG photo, the camera applies the correction that comes along with the lens. This is baked into the JPEG, there is nothing you can do about it in Lightroom. (That is not quite right, you can manually add further barrel or pincushion distortion correction if you want.) Raw is a different story. For RAW files the photo is not corrected for distortion, and the lens profile correction is passed along in the file with the EXIF data. By default Lightroom applies the lens profile correction in the EXIF data. You can turn this off in Lightroom if you want, and see the uncorrected photo. The 18-40mm lens uncorrected has very high barrel distortion at 18mm. Perhaps you might use 3rd party software to correct this, or use the manual correction function in Lightroom.

The 18-40mm lens is designed to be rectilinear after the lens distortion profile is applied. Again, this profile is applied in the camera for JPEG photos, and is applied in post-processing for RAW photos.

For my own understanding I set up a test to see what rectilinear looks like at 18 mm with this lens. I used VIOSO to design a 3:2 test pattern with 96 equal squares in 12 columns and 8 rows. I displayed the test pattern full screen on my monitor, and set up the S1R II on a tripod to photograph the test pattern. I aligned the camera to be centered and the correct distance and perpendicular to the test pattern as best as I could without the aid of a precision alignment setup.

This is the test pattern



This is the JPEG photo in Lightroom

• Panasonic - DC-S1RM2
• LUMIX S 18-40/F4.5-6.3
• 18.0 mm
• ƒ/4.5
• 1/40 sec
• Pattern
• Manual exposure
• ISO 1000

Note the test pattern is not perfectly aligned, but is pretty good. This photo has a small amount of barrel distortion, probably not enough to show in most photos. I printed this out and measured the squares in the middle and at the extremes. Within the limits of this test they are all the sme size. Circles, faces, people would not be distorted under these conditions.


This is the RAW photo in Lightroom

• Panasonic - DC-S1RM2
• LUMIX S 18-40/F4.5-6.3
• 18.0 mm
• ƒ/4.5
• 1/40 sec
• Pattern
• Manual exposure
• ISO 1000

Note this looks the same as the JPEG photo.


This is the RAW photo with distortion correction turned off in Lightroom

• Panasonic - DC-S1RM2
• LUMIX S 18-40/F4.5-6.3
• 18.0 mm
• ƒ/4.5
• 1/40 sec
• Pattern
• Manual exposure
• ISO 1000

Note the high amount of barrel distortion when distortion correction is not applied.


I also took a look at this with DxO, using DxO PureRAW as an add-on in Lightroom. DxO does NOT use the Panasonic EXIF data to correct distortion. Instead they test every lens and derive their own distortion correction profile.

Here is the RAW photo using DxO PureRAW

• Panasonic - DC-S1RM2
• LUMIX S 18-40/F4.5-6.3
• 18.0 mm
• ƒ/4.5
• 1/40 sec
• Pattern
• Manual exposure
• ISO 1000

A first thing to note is the test pattern is a little smaller. DxO always pulls out a little more from wide angle lenses, so the 18-40mm lens at 18mm is a little wider angle with DxO than with the Lightroom RAW conversion. Another thing, there is essentially no barrel or pincushion distortion, so DxO does a better job here than Lightroom. Finally the boxs are all the same size so it is rectilinear.

Again, this testing was for me to see what rectilinear looks like with the 18-40mm at 18mm with Lightroom, and with DxO. I found it useful and I hope it adds a little to this discussion.

 

[Andreas]

Thanks all for the good discussion, and the distinction between lens distortion and perspective distortion.

I'd like to add some comments. ...

...

Note the test pattern is not perfectly aligned, but is pretty good. ... I printed this out and measured the squares in the middle and at the extremes. Within the limits of this test they are all the same size. Circles, faces, people would not be distorted under these conditions.

...

Thanks for these images of flat test charts that are parallel to the sensor plane. It helps to illustrate one of Babylonia's points, which is that you do NOT get stretching when both of these conditions obtain. But both conditions are necessary to avoid stretching: the subject needs to be flat as well as parallel to the sensor. As a result, your conclusions are not fully correct. Yes, a flat circle placed on the edge of the test chart will not appear stretched, but paradoxically a three-dimensional ball (or three-dimensional ball-shaped head) will appear stretched if placed at the same location, for reasons I will attempt to explain. But first, have a look at the image of the tennis racket and tennis balls in Babylonia's first post. The flat squares in the tennis racket are not stretched, but the spherical tennis balls in the same location are stretched, and to a rather extreme degree. So how do we explain this?

Following Babylonia's explanation on the Dutch website (see his link), I would explain it as follows. Let's say that you are facing a wall, and someone places a flat circle on the wall far to your right. When you look at the circle to your right you are observing it at an oblique angle, and so the circle will actually appear oblong, which is a form of perspective distortion. When taking a photo, a rectilinear lens then stretches the image. This provides a second form of distortion, which exactly offsets the first. These two distortions taken together result in the circle at the edge of the frame looking like a circle in the final image. The lens, by stretching the edge of the image, is actually stretching the circle from an apparent oblong back into a round shape! The key is that there are two offsetting perceptual distortions.

So why doesn't this work with a spherical object, such as a tennis ball (or a head)? Because we are missing the first type of distortion! Let's go back to the start, but in this case someone attaches a tennis ball to the wall, far to your right. When you glance over at the ball it will NOT appear oblong, but rather it will appear round, in spite of the oblique angle. No matter how far you move the ball to the right it will always appear round. So when you photographic the scene with a camera parallel to the wall, the stretching caused by the rectilinear lens causes the round-looking ball to be stretched into a distorted shape.

In this case it is better to have two offsetting perceptual distortions, which is what you get when you photograph a flat test chart on a parallel wall. Unfortunately this is not a good physical model of the real world, and is yet another reason why we cannot rely on test charts to reveal the full story.

 

[Babylonia]

The amount of horizontal stretching in the example I posted is more than 20 percent (relative to the panorama), which I wouldn't characterize as "tiny".

It just depends on what you're relating to.
For example, the previously given example < with the stretched tennis balls > is a much higher percentage.

Constructing a panorama is not just a matter of stitching image slices together. Rather, it involves stitching together the image slices and applying a specific projection. In this case I find that a spherical projection works best.

The choice of a spherical projection method you didn't mention in your previous message.
That's why I had my reservations in the different outcome of the result? (As I didn't expect that result).

In my more earlier message I specially wrote:
If one or more of those conditions are different / changed. Not as described above.
The character of the image shall change.

Changing the projection method of the panorama out of three 20mm lens images using spherical projection.
You can NOT compare by a one shot image of the same 20mm lens.
As a one shot image of 20mm lens definitely is not a spherical projection of the image.

When using a super wide angle lens. What we as viewers perceive as a perceptually "correct image".
Can be achieved by keeping the viewing distance relative to the dimensions of a "print" (or screen display)
in exactly the same proportions as the actual image.

Example:
Using a 20mm wide angle lens. Taking a picture from an indoor object / room.
Object distance (focus) = 4.02 meter
The sizes / borders of the taken picture covers an area of 4.80 meter high x 7.20 meter wide

If you print an image 48 cm high x 72 cm wide ---> so 10 times smaller that the real actual object.
And this image is viewed by a viewing distance of only 42 cm = also 10 times smaller that the real actual object.
(So really close viewing distance in relation to the print size).

The perceptual viewing experience when looking to the print is "the same" as the real time object.
(But only as a flat 2D projection). "Without wrong" stretched image parts.
As by this short viewing distance, you are looking perpendicular in the centre part of the print.

But you have to turn your head a little to the left and right to have a better view of the borders and corners.
The looking direction to see these sides / borders become more and more oblique.
(By a gradual progression going from centre to the borders / corners).
By that, "stretched" image areas are seen "shortened". And the image is seen as "normal".

Compare e.g. the "stretched" painted traffic signs on a road when seen from above.
But as a driver of a car, is looking oblique to the road itself, the perceptual viewing experience.
Car drivers experiences these traffic signs as "normal" dimensions.

I used an old - but very handy and precisely DOF calculator programmed by "Paul van Walree".
For above calculations. And coloured the parts of the related data.



-

 

[Babylonia]

Note the test pattern is not perfectly aligned, but is pretty good. This photo has a small amount of barrel distortion, probably not enough to show in most photos. I printed this out and measured the squares in the middle and at the extremes.
Within the limits of this test they are all the same size. Circles, faces, people would not be distorted under these conditions.

No, that is not true.
You only tested for a flat 2D object. These circumstances and what you find out only is true for this flat test pattern.

The outcome is different for 3D objects like faces, (a head is more like a 3D "egg" shape), and 3D people.
(Check the extreme stretched tennis-balls of < my more early example > ).

This effect of "3D" stretched areas can be seen easily when doing "group" portraits, when using a wide angle lens.
Check e.g. regular subjects at fora:
Subject by reddit community



To overcome these stretched persons at the borders, me myself never go wider than using a 40mm focal length as for 24x36mm format.
(In past when using a < Bronica 4.5x6 roll-film camera > - Using a "normal" 75 mm lens).
-

 

[CharlesH]

As a result, your conclusions are not fully correct. Yes, a flat circle placed on the edge of the test chart will not appear stretched, but paradoxically a three-dimensional ball (or three-dimensional ball-shaped head) will appear stretched if placed at the same location

My intent was to only measure and demonstrate the flat field lens distortion of the 18-40 lens at 18mm, and I purposefully did not measure or discuss perceptual distortions, which have already been discussed in the thread. I'm sorry if that wasn't clear.

 

[CharlesH]

No, that is not true.
You only tested for a flat 2D object. These circumstances and what you find out only is true for this flat test pattern.

My intent was to only measure and demonstrate the flat field lens distortion of the 18-40 lens at 18mm, and I purposefully did not measure or discuss perceptual distortions, which have already been discussed in the thread. I'm sorry if that wasn't clear.

 

[Andreas]

It just depends on what you're relating to.
For example, the previously given example < with the stretched tennis balls > is a much higher percentage.


The choice of a spherical projection method you didn't mention in your previous message.
That's why I had my reservations in the different outcome of the result? (As I didn't expect that result).

Actually, as I noted in my correction (a follow-up post), I didn't use a spherical projection in this case, but rather a Panini projection. One of the benefits of using good panorama software is that I have my choice of many different projections and can experiment, using the projection that provides what I perceive as the most natural looking image.

...

Changing the projection method of the panorama out of three 20mm lens images using spherical projection.
You can NOT compare by a one shot image of the same 20mm lens.
As a one shot image of 20mm lens definitely is not a spherical projection of the image.

Of course you can compare the images! That's the entire point of my post. If you are getting stretching at the edges of a one-shot image and want to eliminate it, you can do so by taking multiple images while rotating the camera, and then combining slices of those the images, using any number of possible projections. The result is a very different final image, which needs to be compared to the single shot image to see which is subjectively superior.

When using a super wide angle lens. What we as viewers perceive as a perceptually "correct image".
Can be achieved by keeping the viewing distance relative to the dimensions of a "print" (or screen display)
in exactly the same proportions as the actual image. ...

What I perceive as a perceptually correct image is based on whether the image on the screen matches what I perceive subjectively when I view the scene. Of course this is in part a matter of many factors, including how the human brain translates visual information. Let's say that I am examining an image where there is a mountain on the side of an image that looks fatter to me than it "should", relative to the rest of the scene. In that case I perceive the image as "distorted". I don't really care if it is distorted in an objective sense, I care if I perceive it as distorted. The photographic task then becomes simple. Is there an alternative technique that provides an image that I perceive as subjectively superior? In my opinion the best way to determine this is to actually try a number of alternatives and then to compare the images. It's just that simple.

 

[Andreas]

My intent was to only measure and demonstrate the flat field lens distortion of the 18-40 lens at 18mm, and I purposefully did not measure or discuss perceptual distortions, which have already been discussed in the thread. I'm sorry if that wasn't clear.

I apologize if I seem to have misconstrued your post. It does seem clear that you were focusing on linear distortion, and alternative corrections. That should prove very useful to owners of the lens, particularly as regards the relative performance of DxO. I wouldn't have responded at all except for that one sentence in which you said "faces, people would not be distorted under these conditions". That small error should not distract from the main points in your post.

 

[Pete_W]

I also took a look at this with DxO, using DxO PureRAW as an add-on in Lightroom. DxO does NOT use the Panasonic EXIF data to correct distortion. Instead they test every lens and derive their own distortion correction profile.

Thanks for this Charles. As a DxO PhotoLab user, I was wondering why we have to wait for lens profiles to be provided given that correction info was stored within exif.