Luminous Landscape Forum
Equipment & Techniques => Medium Format / Film / Digital Backs โ and Large Sensor Photography => Topic started by: heinrichvoelkel on January 10, 2014, 03:03:58 pm
-
Maybe this is more a philosophical question, but still I'm going to ask.
Searching for distortion free lenses for digital I was asking myself: Equals distortion correction via software a well corrected lens in quality terms?
What I mean is, with film correcting the lens was important, as later it couldn't be fixed easily. Today you can use software to ease the flaws of lens design (like Hasselblad is using it as a design approach).
So, is our search for the "best" lens (distortion free) only a reminiscence to the way we used to work? With digital pushing pixels is part of the proposition of the medium.
Thanks in advance for your thoughts
-
Maybe this is more a philosophical question, but still I'm going to ask.
Searching for distortion free lenses for digital I was asking myself: Equals distortion correction via software a well corrected lens in quality terms?
What I mean is, with film correcting the lens was important, as later it couldn't be fixed easily. Today you can use software to ease the flaws of lens design (like Hasselblad is using it as a design approach).
So, is our search for the "best" lens (distortion free) only a reminiscence to the way we used to work? With digital pushing pixels is part of the proposition of the medium.
Thanks in advance for your thoughts
Interesting observations. As a tech camera shooter, none of my lens have distortion and I very much like it. But they are designed without having to worry about a mirror box, which helps. (However the newer Rodie lenses have a good deal of distortion that needs to be removed.)
Not the case with Hassy and Mamiya. I would assume designing a lens around a mirror box and than also making it distortion free would make it heavier, more expensive to design, and possibly less sharp. It just may be more advantageous to correct the distortion in post or have it built into the firmware.
For a lens that does not shift, not really an issue to apply this in post. With a lens that shifts, you have to keep track of the shifts for it to work, which is why I prefer fully symmetrical lenses.
-
copy/paste :-)
Interpolation always reduces IQ a bit... this goes for software based lens correction as well.
Then again even the very best lenses regarding distortion (for instance Schneider Digitar symmetrical designs) require a little software correction if you aim at a perfectly corrected and straightened image.
Personally I prefer "distortion free" lenses for 2 reasons:
- the last bit of software correction is not always required (depending on the scene), so in this case you preserve best possible IQ
- less distortion of the lens requires less software interpolation. IMO this helps to get a more homogeneous distribution of "sharpness" all over the image plane (the relation of center and edges is more balanced - while the correction of heavy and complex moustache distortion often leads to softer edges). less software correction also preserves more pixels (less cropping required).
-
Personally, I don't like the Hasselblad "let the optics distort and laterally fringe, just fix it in software" design philosophy. From what I've seen, it works marvellously - as long as you use the 1 or 2 software packages that can make sense of the metadata and perform the correction. But if you want or need to use something else (and I do regularly use non-photo-mainstream products), too bad.
+1 to tho_mas' points as well.
Ray
-
There is a general difference in design philosophy in the area of digital optics from Schneider and Rodenstock.
Especially in wide angle lenses these differences in design create differences in results.
Schneider tends toward a traditional design which yields a lens which will closely match the distance from the nodal point to the sensor, to the focal length quoted. This yields a lens with virtually distortion free image. But the trade-off is that there will be more curvature of field and so generally, a smaller image circle than with Rodenstock's choice of a reverse telephoto design. This because the nodal point is moved back so as to create a greater distance to the sensor and so a larger image is projected onto the image plane and sensor.
But the reverse telephoto design has an unfortunate issue. Justa Schneider has a problem, image circle, Rodenstock's design philosophy also creates an issue: Mustache distortion. If you are shooting a regular shape in the center of your image such as a long horizontal building, you will note that it may exhibit a sort of peak in the center and slowing fall off until near the edge it bends downward quickly.
In a landscape photo it is less of an issue, with architecture, it will require software in post to correct. The reverse telephoto design also results in the lenses being physically larger, and heavier, in most cases than the more traditional designs of Schneider.
So you have to decide, vis-a-vis the subject matter you photograph, which brand you will purchase. Schneider with little or no distortion and great natural color, or Rodenstock with a generally larger image circle and a bit higher saturation. Both are incredibly sharp so that is a
non issue. Both produce superb images...
Rod
-
I have pretty good knowledge of the Hasselblad lens corrections and the guys who design them.
I'd like to say it's absolutely NOT a case of let's design a sub-standard lens and fix it later. What they do is something like this:
They discuss if they should be trying to address an issue issue with say, multi-coatings (chromatic aberrations) or can they do a better job by moving this channel (say the a of lab) a few pixels based on the distortion map of this lens. Which will give them a better result? Yes cost is a factor as the lens will have to hit a particular price point but it is not true to say that software corrections are used instead of good design the corrections are just another tool in the designers arsenal.
As an aside get someone to show you the digital corrections for distortion created by movements with the HTS system, nothing short of amazing IMO.
-
Hi,
Multi coating has absolutely nothing to do chromatic aberration. That is pure nonsense! Sorry!
I feel it is OK to fix vignetting and distortion in software, but I would strongly suggest that chromatic aberration should be corrected in optical design.
Most lenses do suffer from chromatic aberration, however, and it is obviously better to fix chromatic fringing in software than not fixing at all.
Best regards
Erik
I have pretty good knowledge of the Hasselblad lens corrections and the guys who design them.
I'd like to say it's absolutely NOT a case of let's design a sub-standard lens and fix it later. What they do is something like this:
They discuss if they should be trying to address an issue issue with say, multi-coatings (chromatic aberrations) or can they do a better job by moving this channel (say the a of lab) a few pixels based on the distortion map of this lens. Which will give them a better result? Yes cost is a factor as the lens will have to hit a particular price point but it is not true to say that software corrections are used instead of good design the corrections are just another tool in the designers arsenal.
As an aside get someone to show you the digital corrections for distortion created by movements with the HTS system, nothing short of amazing IMO.
-
Hi,
I tend to agree with both tho_mas and Ray on the issue, but I feel this type of correction is well proven, let's not forget that the transformations are also used in panoramic software.
The fringing is in my humble opinion a greater problem. I am pretty sure that both resolution and edge contrast is lost. But, most lenses have chromatic aberration and it is better to fix them in software than not at all.
I checked the MTF curves and Distortion curve on the 4/28 H lens just as an example, I don't think chromatic aberration is a great problem on that lens judging from the MTF of that lens. Chromatic aberration is often noticable as a nearly linear drop of radial MTF with increasing offset from optical axis. Distortion is pretty bad, however.
Best regards
Erik
Personally, I don't like the Hasselblad "let the optics distort and laterally fringe, just fix it in software" design philosophy. From what I've seen, it works marvellously - as long as you use the 1 or 2 software packages that can make sense of the metadata and perform the correction. But if you want or need to use something else (and I do regularly use non-photo-mainstream products), too bad.
+1 to tho_mas' points as well.
Ray
-
It is quite easy to visualise the lens with and without corrections in phocus, the corrections can be clicked on and off. I would say that the lenses are already pretty good before corrections. You should try to see what happens with some ยต4/3 or P&S lenses without corrections.
I any cases, everything is a compromise in optics. It is possible to construct better lenses, reduce distortion and chromatic aberrations, etc... But the lenses will be bigger, heavier and more expensive.
-
Maybe this is more a philosophical question, but still I'm going to ask.
Searching for distortion free lenses for digital I was asking myself: Equals distortion correction via software a well corrected lens in quality terms?
Hi Heinrich,
The answer is no, they are different, but both can affect image quality negatively. The question then becomes, which deteriorates the image more, the residual lens aberrations or the software resampling. In addition, it can be a combination of improvements, some are better done in lens design and others better in software, thus producing fewer design compromises for the lens designers, and a better combined result.
I do not hope that it results in a; sloppy-design, we'll fix it in post, attitude. And as e.g. the OTUS and Sigma Art lenses show, it helps to get both the lens design and the postprocessing right, but at a cost.
There is also a difference how the software correction is implemented. When it is a part of the Raw conversion process, the potential gains can be much better than post-processing of an already rendered RGB file. For example, correcting Lateral Chromatic Aberration, can sometimes be done before demosaicing, which will allow higher resolution and more accurate color conversions.
Cheers,
Bart
-
none of my lens have distortion and I very much like it.
I was not aware that a perfect lens had been designed and made. ;)
-
I was not aware that a perfect lens had been designed and made. ;)
Haha. I should have said that my SK 35mm (my go to lens) has, at most, 0.27% barrel distortion, depending on how far away your are from center. So virtually none, or none that I can see, even with a grid over top.
-
Rod:
Thanks for the info on the Schneiders and Rodenstocks.
What's interesting to me over the distortion is how the Schneider 90mm image circle vs Rodenstock holds up to shifting. The Schneider 90mm on the SK35 is only good to about 8mm of shift on a 60MP or larger back, where as the Rodenstock 32 or 40mm can easily go to 15mm and would go further if not for the internal vignetting disc that Rodentock places inside their lenses to show the visual limit of the image circle ( or so I assume ). If not for the disc the 40mm would easily make 20mm of shift.
Paul Caldwell
-
Maybe this is more a philosophical question, but still I'm going to ask.
Searching for distortion free lenses for digital I was asking myself: Equals distortion correction via software a well corrected lens in quality terms?
What I mean is, with film correcting the lens was important, as later it couldn't be fixed easily. Today you can use software to ease the flaws of lens design (like Hasselblad is using it as a design approach).
So, is our search for the "best" lens (distortion free) only a reminiscence to the way we used to work? With digital pushing pixels is part of the proposition of the medium.
Thanks in advance for your thoughts
a simple illustration of software optics correction that ACR/LR (for example) does for m43 optics...
http://imagizer.imageshack.us/v2/800x600q90/22/gf6z.jpg (http://imagizer.imageshack.us/v2/800x600q90/22/gf6z.jpg)
(http://imagizer.imageshack.us/v2/800x600q90/22/gf6z.jpg)
you see the noise banding ? that's by design (that is what software correction does) as noted by Eric Chan of Adobe Labs... that is about geometry correction (not about LaCA or LoCA)
-
a simple illustration of software optics correction that ACR/LR (for example) does for m43 optics...
you see the noise banding ? that's by design (that is what software correction does) as noted by Eric Chan of Adobe Labs... that is about geometry correction (not about LaCA or LoCA)
Regarding the banding. It does depend on the sensor aka quality of the sensor in matters of being prone to banding in general? Or is this a general observation.
-
thank you all for the interesting observations and opinions.
I didn't mean to point fingers at Hasselblad, actually I think their approach makes sense for the system they designed.
In general I'm yet not convinced that distortion correction as example for interpolation is so different from correcting CA. If you start to push the pixel around, it makes no difference โ you mess with the original captured picture. What I'm thinking is more in the direction of the obvious in the world of digital capture. Does it make a difference if there is interpolation, because the capture is 1 and 0 anyway and no concrete and physical evidence of the capture does exist ( unlike with film).
I hope my thoughts make sense?!
-
Regarding the banding. It does depend on the sensor aka quality of the sensor in matters of being prone to banding in general? Or is this a general observation.
if you do saturate your sensor quite (proper gain /what is known otherwise as ISO/ & good exposure /time & aperture/) well and don't push the contrast a lot you mostly don't see it... busy/irregular background and color instead of BW also helps... but it is always there...
PS: that image was shot with Panasonic 35-100/2.8 @ 85mm tests shows that barrel distortion is โ1.90% in the middle of the range (~60mm) and to 0.65% at 100 mm... so you might assume that @ 85mm is shall be a no big deal... not so if you stress the data (shot was @ ISO6400, poor light and contrast was made high during conversion)... if you don't push it you don't see it w/ naked eye.
-
Hi,
An illustration without context is pretty meaningless, I see a distorted pattern but where does it come from, what does the real image look like?
Best regards
Erik
a simple illustration of software optics correction that ACR/LR (for example) does for m43 optics...
http://imagizer.imageshack.us/v2/800x600q90/22/gf6z.jpg (http://imagizer.imageshack.us/v2/800x600q90/22/gf6z.jpg)
(http://imagizer.imageshack.us/v2/800x600q90/22/gf6z.jpg)
you see the noise banding ? that's by design (that is what software correction does) as noted by Eric Chan of Adobe Labs... that is about geometry correction (not about LaCA or LoCA)
-
Here's maybe a more tonally representative m43 Panasonic example. This is from the GX7 camera and 20mm lens. First a downsample from the corrected-in-camera JPEG, then an uncorrected downsample from the RAW file using PhotoRAW on my iPad. In the second version the framing exaggerates the barrel distortion on the right side, but you can see it as well on the left.
-Dave-
-
and the Sony RX100:
-
I see a distorted pattern but where does it come from
as I noted it comes from software correction of geometric distortion when noise ("grain") is "banded" in such "clusters" as a result of the transform applied (example just makes that more visible), if curious then you might want to visit http://forums.adobe.com/community/cameraraw and find the relevant topic with the direct answer from Eric Chan of Adobe Labs... that is not unique to ACR/LR - SilkyPix does the same.
-
@ reply #9
Hi Bart,
> There is also a difference how the software correction is implemented. When it is a part of the Raw conversion process, the potential gains can be much better than post-processing of an already rendered RGB file. For example, correcting Lateral Chromatic Aberration, can sometimes be done before demosaicing, which will allow higher resolution and more accurate color conversions.
Could you specify software which does this? And/or give some general hints, what you would recommend for CA correction?
Good light - Hening.
-
Thanks!
I have not seen something similar but I can guess a couple of reasons for it, like shooting low ISO and not using corrections very often.
Good to be aware of the problem!
Best regards
Erik
as I noted it comes from software correction of geometric distortion when noise ("grain") is "banded" in such "clusters" as a result of the transform applied (example just makes that more visible), if curious then you might want to visit http://forums.adobe.com/community/cameraraw and find the relevant topic with the direct answer from Eric Chan of Adobe Labs... that is not unique to ACR/LR - SilkyPix does the same.
-
@ reply #9
Hi Bart,
> There is also a difference how the software correction is implemented. When it is a part of the Raw conversion process, the potential gains can be much better than post-processing of an already rendered RGB file. For example, correcting Lateral Chromatic Aberration, can sometimes be done before demosaicing, which will allow higher resolution and more accurate color conversions.
Could you specify software which does this? And/or give some general hints, what you would recommend for CA correction?
Hi Hening,
RawTherapee can do it for Lateral CA, at the Raw conversion stage. It also offers traditional post-demosaicing LaCA correction for the occasion it doesn't work that well at the Raw level (depends of accurate auto-detection and complexity, I assume). I image that DxO might do some distortion correction at a very early stage, but I don't know for sure how that's implemented.
I think most commercial converters take the old-fashioned approach of fixing things after demosaicing, but that may also be connected to the fact that very few converters work in floating point precision (like RawTherapee does). Being able to calculate at sub-pixel accuracy and with huge dynamic range differences takes a toll on memory requirements, but also opens up a lot of potential for better algorithms.
There are some Astrophotography related applications that do use floating point number brightness calculations e.g. PixInsight, but their tools are somewhat specifically tuned for those kind of images.
Cheers,
Bart
-
Thank you, Bart.
-
Rectilinear distortion varies with the distance from subject to camera. For something like architectural photography, this can sometimes be a quite significant consideration, depending upon how critical you want to be. No software can completely correct for different amounts of distortion in different parts of the image, in all situations, especially with global corrections.
-
No software can completely correct for different amounts of distortion in different parts of the image, in all situations, especially with global corrections.
Hasselblad's Phocus software does precisely that. Corrections are based on the design of the lens (Hasselblad having designed the lens) and take into account aperture and camera to subject distance. As an aside the lens corrections when using the HTS are quite amazing as the software knows exactly how much shift/tilt has been employed and corrects accordingly.
Also thanks for the correction above re multi-coatings I was just trying to come up with an example of what can be done in design vs what can be done in post, for example post processing cannot fix a lens that is soft at the edges so this needs to be addressed at design. Distortion however, can be very effectively corrected in post (assuming you have the right data) so having lens corrections gives designers more options as to where to devote their resources.
Hope that makes sense.
Nick-T
-
Hi,
I am quite impressed by Hasselblad's way of thinking out of the box. I have flexbody myself. Regarding the software corrections, I would say that is OK.
Best regards
Erik
Hasselblad's Phocus software does precisely that. Corrections are based on the design of the lens (Hasselblad having designed the lens) and take into account aperture and camera to subject distance. As an aside the lens corrections when using the HTS are quite amazing as the software knows exactly how much shift/tilt has been employed and corrects accordingly.
Also thanks for the correction above re multi-coatings I was just trying to come up with an example of what can be done in design vs what can be done in post, for example post processing cannot fix a lens that is soft at the edges so this needs to be addressed at design. Distortion however, can be very effectively corrected in post (assuming you have the right data) so having lens corrections gives designers more options as to where to devote their resources.
Hope that makes sense.
Nick-T
-
Basically what's going on in the image with the pattern artifacts is that an image that an evenly distributed amount of noise is being warped (resampled). Doing so results in the (previously even) noise becoming unevenly distributed. This will become especially apparent if you leave the noise present and use strong sharpening settings.
-
Hasselblad's Phocus software does precisely that. Corrections are based on the design of the lens (Hasselblad having designed the lens) and take into account aperture and camera to subject distance. As an aside the lens corrections when using the HTS are quite amazing as the software knows exactly how much shift/tilt has been employed and corrects accordingly.
Also thanks for the correction above re multi-coatings I was just trying to come up with an example of what can be done in design vs what can be done in post, for example post processing cannot fix a lens that is soft at the edges so this needs to be addressed at design. Distortion however, can be very effectively corrected in post (assuming you have the right data) so having lens corrections gives designers more options as to where to devote their resources.
Hope that makes sense.
Nick-T
It is impossible for the software to know what all the (possibly infinite) subject-to-camera distances are in a particular image. I repeat, rectilinear distortion varies with the distance of the subject to the camera. So, in the same image, an object that is, say, 5 feet from the camera can exhibit a different amount of distortion from one that is 50 feet away. How much of difference will depend upon the amount of inherent distortion in the lens's design. Furthermore, distortion cannot always be corrected in post, or, if it can, can sometimes require extreme amounts of selective retouching. Just trying to move things around with Photoshop Transform/Warp might correct one problem but create another one in the process by distorting nearby objects that do not exhibit the same amount of distortion as the area you are trying to fix. Architectural photographers are very sensitive to this kind of thing.
-
It is impossible for the software to know what all the (possibly infinite) subject-to-camera distances are in a particular image.
Hi David
This data does actually get passed back to Phocus, not to the nearest millimetre for sure but the focus distance is recorded and used for the corrections.
Nick-T
-
Hi David
This data does actually get passed back to Phocus, not to the nearest millimetre for sure but the focus distance is recorded and used for the corrections.
Nick-T
Can you cite some authoritative documentation to support your contention (preferably with a link)?
Your argument still does not sufficiently address the issue I am discussing. Simply adjusting the correction somewhat according to the location of the plane of focus is not sufficient to address this issue. The software would need to address variations in the amount of distortion for objects not at the plane of focus.
-
It is impossible for the software to know what all the (possibly infinite) subject-to-camera distances are in a particular image.
Hi David
This data does actually get passed back to Phocus, not to the nearest millimetre for sure but the focus distance is recorded and used for the corrections.
Nick-T
Nick - surely you mean *a* distance gets passed back - the distance that the camera or user chose as the subject plane to be focused on. David's right: there's no way that the camera can simultaneously measure *distances* all over the image to foreground, background and "side-ground" details. (It would need a grid of thousands of AF sensors covering the entire focal plane. Or else, a complete redesign to employ the sort of wavefront phase-recording technology used in the Lytro camera and similar prototypes).
Consequently, Hasselblad's corrections for distortion can only be optimal for the nominal focused distance.
I don't know if the situation is any better for a highly-corrected lens without software corrections, though. It too will presumably have variable residual distortion for objects at different distances from the camera.
Ray
-
Hi,
I guess that most lenses having very low distortion are symmetrical designs so they may be less dependent on distance.
In my view it makes sense to make a lens with a small and simple low order distortion and reduce it in post instead of creating a lens with a small but complex high order distortion.
Best regards
Erik
I don't know if the situation is any better for a highly-corrected lens without software corrections, though. It too will presumably have variable residual distortion for objects at different distances from the camera.
Ray
-
Hi,
I guess that most lenses having very low distortion are symmetrical designs so they may be less dependent on distance.
In my view it makes sense to make a lens with a small and simple low order distortion and reduce it in post instead of creating a lens with a small but complex high order distortion.
Best regards
Erik
I am no expert in optics or lens design, but I believe it is true that symmetrical lens designs will generally tend to have less rectilinear distortion. However, this is not to say that other lens designs cannot have very low levels of such distortion as well. I believe the biggest problem with such distortion occurs with retrofocus designs for wideangle lenses for reflex cameras. Conversely, I believe that symmetrical lens designs tend to have a higher amount of vignetting, compared with non-symmetrical designs at the same focal length, image circle and aperture. With some wideangle lenses of symmetrical or near symmetrical design, such as the 90mm Super Angulon, you need to use a special graduated ND filter to compensate for the high degree of vignetting.
Also, I don't mean to suggest that lenses with a high degree of rectilinear distortion cannot be used successfully (with the appropriate digital distortion correction) for demanding applications, just that you need to be more careful how you use them and recognize the limitations. Furthermore, sometimes the benefits of using a zoom lens will outweigh some limitations.
-
Nick - surely you mean *a* distance gets passed back -
Yes that's what I meant thanks Ray.