What are the consequences of in-camera lens correction?

This is not about the pros and cons of in camera corrections and it is designed into most mFT lenses by default which is ok as it helps keep size & weight of lenses down while combing to deliver good end results. I would just like to understand what is going on with it. These examples are from the Olympus 12-40mm { on the excellent E-M1} . My questions are the area highlighted in the uncorrected file which is sacrificed to distortion correction, when the file is resized back to the original size is it done by simple re-sizing or does it involve pixel stretching or other processes Regarding the resulting loss of the AOV do you think this is accounted for in the lens design in other words is the lens wider than 12mm with the post correction image being the actual stated AOV.

RAW files source

http://www.photographyblog.com/previews/olympus_om_d_e_m1_photos/

Settings used

ACR { NR/sharpening set to 0 , defaults for everything else }

Photo Ninja { { NR/sharpening/distortion control all turned off}

F/2.8 12mm { FF AOV 24mm }

vs

F/2.8 12mm { FF AOV 24mm }

Vs

It is not just wide open that corrections are going on this is F/11 12mm { FF AOV 24mm }

I assume it is difficult / impossible to tell what other processes are going on behind the scenes in the camera or in Photoshop. Looking in detail at some of the very high ISO files I suspect that either the Olympus RAW file is instructing ACR to add some kind of NR or ACR does this behind the scenes. Simply suggesting this in the mFT forum typically results in a barrage of abuse and emphatic declarations this not to be the case.

As a rule I am a happy ACR RAW shooter and do not get too involved in the technical nitty-gritty. Though I was aware of the software aspect of mFT lens performance { I have been a mFT user since 2009 } I was a bit surprised by its extent even as here on what is considered an excellent lens when I looked at RAW software that lets you see the uncorrected images. . There is so much bias in the respective camera format forums that such discussions are typically unhelpful with a high potential for conflict. So I thought posting here in a forum with more than its fair share of technical gurus/geniuses would be my wisest option. It is perfectly possible that I am talking nonsense and I have no problem being told so by people who actually know what they are talking about.

100% detail crop form ACR { NR/sharpening set to 0 , defaults for everything else } vs 100% crop from Photo Ninja { { NR/sharpening/distortion control all turned off}. Using a 6400 ISO file

f/5.6 12mm { FF AOV 24mm } 6400 ISO 100% crop

NR?

Please look at the original files as DPreview downsizing is far from perfect. Thank you for taking the time to read my post . And I look forward to some input from the informed crowd that posts in this forum. I would also like to apologise up front if this post draws in any of the mFT diehards , I will ignore them as polite dialogue is not possible

cheers

Jim

Tom Axford wrote:

No one else has replied so far, so I will make a few observations which may hopefully be helpful, although I do not know the full answer to your question.

Thanks for replying Tom

First, it doesn't really matter how distortion correction takes place. Comparisons of cameras by DPR and other people are almost always done on the final image (after lens corrections have been applied). I do not have an E-M1, but it is undoubtedly a very good camera judging it on the final images produced. Unless you intend to take the lens off the camera and use it for a totally different purpose, it doesn't matter how much optical distortion the lens has. What matters is the quality of the final image.

Comparisons by DPreview are done on the default RAW files { they seem to use ACR } which on mFT and perhaps other mirrorless cameras . Have already had distortion correction etc which gives an impression that somehow these lenses are better than they physically are. The claims about the wonders of mFT lenses { some are wonderful no doubt } which are based on an optical+ software correction routine should be taken with a pinch of salt, or at least only compared to results from optical lenses that have had the same opportunity for distortion and CA corrections in post.

Second, it is apparent from your posted images that the images produced (after lens corrections) by ACR and PN are slightly different in size or scale. If you took a jpeg straight from the camera it may be slightly different again (you could easily check). It appears that each processing program has its own way of converting the raw file to a jpeg and they may apply the lens corrections in slightly different ways. I presume both will produce images of 4608x3456, as the camera spec requires, but you could check that yourself.

I used RAW samples and the file size in ACR is 4608 x 3456 15.9mp in PN the file size is 4640 x 3472 16.1mp so there is a small size difference.

Another observation - the specs for the E-M1 say the sensor has 16.85 million pixels, but the image produced is 4608x3456, which is 15.93 million pixels. Presumably the extra pixels provide extra image data so that distortion correction can be applied without having to resize the image upwards.

That was my initial thought, however when you look at some wide angle samples there appears to be some degree of pixel stretching going on ,look at the room door in this sample

vs

However, some MFT lenses produce optical distortion of around 4% or 5% which would require even more than 16.85 million pixels to correct without resizing the whole image, so I would guess that some resizing is needed to correct the most extreme distortion.

That is also my conclusion .

I think if you really want to know in detail how lens corrections are applied, you may have to ask Olympus and my guess is that they will not tell you. For whatever reason, manufacturers often seem to keep this sort of information out of the public domain. I agree that it would be nice to know (out of curiosity), but this sort of technical information seems very hard to come by (particularly from the Japanese manufacturers, some companies such as Zeiss seem to be more forthcoming).

That is true the info is hard to come by , though there are some very technically gifted folk on this forum and perhaps some RAW software guru will help me out.

_sem_ wrote:

I find the concept of "obligatory" corrections funny. I don't mind if they're supplied as defaults, but... As you've noticed, often you'd get more of the image without them. Further, wideangle images often look better with a bit of barrel (faces in the corners...). The baked-in corrections may be totally out of place when using macro close-up lenses ore extension tubes. And double corrections may cause degradation when doing specifc processing, for instance pano software.

Yes I was wondering if you could achieve better results using the RAW files in software that doesn't use the baked in instructions. Though as  essentially an almost totally ACR user I was largely in the ignorance is bliss scenario till I was tempted down the dark path of non-supported RAW software.

knickerhawk wrote:

jim stirling wrote:

This is not about the pros and cons of in camera corrections and it is designed into most mFT lenses by default which is ok as it helps keep size & weight of lenses down while combing to deliver good end results. I would just like to understand what is going on with it. These examples are from the Olympus 12-40mm { on the excellent E-M1} . My questions are the area highlighted in the uncorrected file which is sacrificed to distortion correction, when the file is resized back to the original size is it done by simple re-sizing or does it involve pixel stretching or other processes

It looks to me like Photo Ninja is doing distortion control also. If you look at a 12mm m4/3 image that really has been processed with no distortion correction, you'll see very obviously convex (barrel) or concave (pincushion) edges. Use a processor like RPP to see this.

How much worse can it get plus I believe that RPP is only for the accursed mac OS

It's well known that the m4/3 lenses usually capture an AOV that is wider than the stated one and that the lens-aware processors like ACR then crop out the excess perimeter (after applying the distortion corrections). There seems to be excess "breathing room" built into some of the lens corrections, meaning that you can manually correct the distortions and get straight edges and still have a bit extra yet that usable extra gets cropped by ACR anyway because Oly/Panny has designed in some extra margin. Your PN example below might actually be illustrating that phenomenon by not doing the full crop down to the AOV dictated by 12mm. I expect that you would find that the PN version is actually between 10-11mm. Consider it a bonus like an unlisted song at the end of a CD...

Where does the "well known" data that the mFT lenses capture wider AOV than the stated one come from it wasn't mentioned in the responses when I initially asked this exact question in the mFT forum. Contrary to what you may be imagining I am doing ,all I am looking for is a clear concise accurate explanation of the process and wondering if as you suggest it would be possible to correct the distortions in other software that would leave you with a wider AOV at wide angle every mm can make a difference.

A lot of what motivates me is the declaration of how amazing mFT lenses are when compared to other lenses in other systems { most of which can be easily improved upon by corrections in post} . The only apples to apples comparison is lens + software correction = final image which many mFT lenses benefit from.

Regarding the resulting loss of the AOV do you think this is accounted for in the lens design in other words is the lens wider than 12mm with the post correction image being the actual stated AOV.

Yes, as explained above. Of course, it's "accounted for" in the lens design and the associated Op-codes provided to the raw processors.

RAW files source

http://www.photographyblog.com/previews/olympus_om_d_e_m1_photos/

Settings used

ACR { NR/sharpening set to 0 , defaults for everything else }

Photo Ninja { { NR/sharpening/distortion control all turned off}

F/2.8 12mm { FF AOV 24mm }

vs

F/2.8 12mm { FF AOV 24mm }

Vs

It is not just wide open that corrections are going on this is F/11 12mm { FF AOV 24mm }

I assume it is difficult / impossible to tell what other processes are going on behind the scenes in the camera or in Photoshop. Looking in detail at some of the very high ISO files I suspect that either the Olympus RAW file is instructing ACR to add some kind of NR or ACR does this behind the scenes. Simply suggesting this in the mFT forum typically results in a barrage of abuse and emphatic declarations this not to be the case.

The so-called "barrage of abuse" is due to the fact that you persist in repeating these claims without carefully equalized samples and associated measurements. Moreover, your speculation flies in the face of DXO's analysis, which has never found that the m4/3 raw files contain "cooked" data.

The barrage of abuse { have you noticed the number of vanishing posts in the forum a result of complaints } is due to the actions of a few rampant fanboys one of whom has been involved in abusive interactions with multiple other posters. Just as in this thread I explain exactly what I have done to the RAW file and more importantly provide links to the RAW files. As for my suspicions about NR in ACR even when it is turned off in the slider , I have since read suggestions that ACR may be adding some level of NR to RAW files not just those from mFT even when turned off.

Unless you use lesser known far less common RAW convertors the mainstream options will honor the in-camera corrections by means of an instruction "header" op-code or whatever it is called.Thus if you use any of the most popular { by a mile } RAW processors you will never see the non-corrected image so for most people who do not use the more unusual convertors , what exactly is the difference between a cooked file and what they will end up with ?

I have been using mFT since 2009 and will continue to do so , despite having a full FF kit , mFT accounts for the majority of my photography. I appreciate that there are some trolling posters in the forum who do not own any mFT gear. I am not one of them and I am only looking for more understanding of what can be a complex issue

As a rule I am a happy ACR RAW shooter and do not get too involved in the technical nitty-gritty. Though I was aware of the software aspect of mFT lens performance { I have been a mFT user since 2009 } I was a bit surprised by its extent even as here on what is considered an excellent lens when I looked at RAW software that lets you see the uncorrected images. . There is so much bias in the respective camera format forums that such discussions are typically unhelpful with a high potential for conflict. So I thought posting here in a forum with more than its fair share of technical gurus/geniuses would be my wisest option. It is perfectly possible that I am talking nonsense and I have no problem being told so by people who actually know what they are talking about.

I know you haven't paid attention to me or Anders or others in the m43 forum, but maybe some "unbiased" observer on this forum will chime in and explain to you how DXO tests for this and why that's likely to be a better answer than what you've convinced yourself based on trying to compare results from different raw processors.

Trusting an analysis based solely on a trawl through the various options looking for and only referencing the one that supports your argument is the definition of biased. There are far more seriously well informed posters in this forum than in mFT. I have read numerous of Anders epic posts . I have stopped doing so and added him to my ignore list for personal abuse. His tendency to use arbitrary definitions such as his take on efficiency which was comprehensibly taken apart on this forum by genuinely competent posters such as Bobn. It is a tad ironic quoting DXO as a reference source when it has received such unwelcome attention in from numerous mFT forum posters.

Also please bear in mind that those who don't shoot m43 may be "unbiased" in one respect but they are also far less likely to have spent time and effort doing careful comparisons of m43 files in various raw processors. Why would they? There's nothing of direct value to them in understanding how raw processors are handling m43 files. On the other hand, some of us who are fully invested in the format have spent a lot of time comparing, tweaking, trying different processors and observing very closely and carefully what's happening in each of those processors.

It is exactly the possibility of the experimenter-expectancy effect from avid supporters of the system I wish to avoid. Having zero personal experience with FF digital does not seem to hinder certain mFT fanboys from pontificating in derogatory condemnation of the format. My questions here are in essence about RAW software and behind the scenes processing going on considering the very high level of experience and technical accomplishments of key posters in this forum I fully expect that they will have a very comprehensive understanding of the processes involved. Whether they respond to my what is no doubt mundane request compared to some of the high level discussion that go on here is a different matter

Please look at the original files as DPreview downsizing is far from perfect. Thank you for taking the time to read my post . And I look forward to some input from the informed crowd that posts in this forum. I would also like to apologise up front if this post draws in any of the mFT diehards , I will ignore them as polite dialogue is not possible

That was uncalled for.

No it wasn't and you are already here demonstrating exactly why I chose to avoid posting this in the mFT, I am sure others will be along just shortly . It is simply not possible to have any discussion in the mFT forum that is not a song of praise for the system as you are invariably inundated with defensive often abusive posts by a small group of posters . You are a reasonable and typically non-confrontational non-abusive poster, and I replied to you as such. Unfortunately this is not the case with certain others and I will not be responding to any posts they may make.

The_Suede wrote:

First, consider what the distortion correction is actually doing...

It applies a non-uniform resampling grid over the original image. For barrel-type distortion, image data that is close to image center is pulled in (downsampled) and image data that is out in the corners are pushed out (upsampled). For pin-cushion type distortion, you get the opposite; upsampling in the middle, downsampling in the corners.

There are two parameters controlling the quality of the end result:

1. The quality of the resampling (scaling) algorithm
2. The quality of the distortion map

Those can of course differ between different software solutions, but that's a given. Adobe's ACR isn't optimal - but it's pretty good, actually.

And then as a more general answer; when you resample the original data, you do of course change the MTF behavior over the image field. In areas where you upsample, you get worse "sharpness". Where you downsample, you get better "sharpness". At least if the resampling algorithms are doing what they should be doing... But as long as the resampling loss is low relative to the original image sharpness, then sharpness loss is negligible.

Ideally you should keep upsampling to a minimum, while losing as little total px resolution as possible. You have to lose SOME of the active sensor area, if you want straight outer edges in your end result - some parts of the original, either the corner parts or the centers of the sides, have to be cropped away.

It's up to the electroptical engineering of the manufacturer to balance these two against each other, and to the general lens construction targets. If they feel that they can get better (or same, but cheaper!) results by allowing for some distortion that is to be corrected in software, then we'll have to trust their judgement.....

As for what the correction does to FoV and so on - that's a totally different, and a lot more complicated question... You always lose FoV by applying a distortion correction. But exactly how much is a hard question, it's much easier to verify that by pure practical experimentation...

Thank you for your considered response.

Steen Bay wrote:

jim stirling wrote:

How much worse can it get

On compacts it can sometimes be worse. If shooting a testchart like in the example below it's easy to see how much the original image must be cropped and stretched/upsampled in order to correct the distortion.

Yes ,I did see that when I was looking at buying the LX100 or RX100 { though I finally ended up with the FZ1000 ]  it was  not pretty . You can see why major compromises have to be made in such physically tiny optics,

If the original image doesn't have severe vignetting and/or corner softness, then it can maybe sometimes be used as it is (if shooting for example nature/landscapes). A bit like having an optional built-in semi fish-eye lens with a wider FoV.

http://www.dpreview.com/forums/post/54790327

I played about with some of the uncorrected RAWs from the LX100 and i felt that it was possible in some circumstances to do a bit better than the in-built  instructions and pinch an extra mm or 2

knickerhawk wrote:

jim stirling wrote:

knickerhawk wrote:

jim stirling wrote:

This is not about the pros and cons of in camera corrections and it is designed into most mFT lenses by default which is ok as it helps keep size & weight of lenses down while combing to deliver good end results. I would just like to understand what is going on with it. These examples are from the Olympus 12-40mm { on the excellent E-M1} . My questions are the area highlighted in the uncorrected file which is sacrificed to distortion correction, when the file is resized back to the original size is it done by simple re-sizing or does it involve pixel stretching or other processes

It looks to me like Photo Ninja is doing distortion control also. If you look at a 12mm m4/3 image that really has been processed with no distortion correction, you'll see very obviously convex (barrel) or concave (pincushion) edges. Use a processor like RPP to see this.

How much worse can it get plus I believe that RPP is only for the accursed mac OS

It's well known that the m4/3 lenses usually capture an AOV that is wider than the stated one and that the lens-aware processors like ACR then crop out the excess perimeter (after applying the distortion corrections). There seems to be excess "breathing room" built into some of the lens corrections, meaning that you can manually correct the distortions and get straight edges and still have a bit extra yet that usable extra gets cropped by ACR anyway because Oly/Panny has designed in some extra margin. Your PN example below might actually be illustrating that phenomenon by not doing the full crop down to the AOV dictated by 12mm. I expect that you would find that the PN version is actually between 10-11mm. Consider it a bonus like an unlisted song at the end of a CD...

Where does the "well known" data that the mFT lenses capture wider AOV than the stated one come from it wasn't mentioned in the responses when I initially asked this exact question in the mFT forum. Contrary to what you may be imagining I am doing ,all I am looking for is a clear concise accurate explanation of the process and wondering if as you suggest it would be possible to correct the distortions in other software that would leave you with a wider AOV at wide angle every mm can make a difference.

Here's one of numerous threads on the topic. There are other discussions on the issue scattered around the internet. It should also be obvious to you from looking at the sample image you processed in PN and ACR. If the PN rendering is corrected sufficiently for your taste (or if a little extra manual correction gets you there), then you should be a happy camper. If you want to squeeze out the most uncropped pixels from your WA mFT lens-based shots, don't bother using ACR because ACR (like the Oly's own raw processor) overcompensates and not only corrects the distortion but also crops a bit extra.

I know there are plenty of chats about it that is rather different from definitive proof . If by using different RAW processors correction scenarios you can squeeze an extra mm or 2 that is certainly worth looking into. The price difference between wider UWA lenses compared to not so wide is very substantial. Though the Panasonic is better built and has a constant aperture range the 7-14mm costs £800 here in the UK while the Olympus 9-18mm costs roughly half that price. If you could get close to a 7mm AOV with reasonable results after external lens corrections that would be quite interesting. I do not do enough UWA with mFT to make it worth buying a more expensive option { for UWA I shoot with my D810 + 14-24mm} . However if you could achieve reasonable results the 9-18 would certainly be worth considering .

A lot of what motivates me is the declaration of how amazing mFT lenses are when compared to other lenses in other systems { most of which can be easily improved upon by corrections in post} . The only apples to apples comparison is lens + software correction = final image which many mFT lenses benefit from.

If I understand you correctly, you're asserting that mFT lenses are "cheating" because of the auto-application of software correction. That's a tired argument that probably won't generate a lot of interest on this forum. If you want to stir that pot, I suggest you head back over to the mFT forum.

No,I am not the in-camera corrections are part of the mFT standard and as I have repeated multiple times. I am ok with that as it in theory allows for smaller lighter better performing lenses. If every mFT lens was optically corrected especially the wide angle options , they would be larger heavier and in all likelihood more expensive. However the constant claims about the superiority of mFT lenses when compared to lenses that have not benefited from software enhancements in other systems is not reasonable a mFT lens results is the result of a combination of optical and software correction { some of which is pretty extensive}.

The so-called "barrage of abuse" is due to the fact that you persist in repeating these claims without carefully equalized samples and associated measurements. Moreover, your speculation flies in the face of DXO's analysis, which has never found that the m4/3 raw files contain "cooked" data.

The barrage of abuse { have you noticed the number of vanishing posts in the forum a result of complaints } is due to the actions of a few rampant fanboys one of whom has been involved in abusive interactions with multiple other posters. Just as in this thread I explain exactly what I have done to the RAW file and more importantly provide links to the RAW files. As for my suspicions about NR in ACR even when it is turned off in the slider , I have since read suggestions that ACR may be adding some level of NR to RAW files not just those from mFT even when turned off.

The crop you posted didn't show less noise in the ACR version compared to the PN version and you haven't offered any links or details to otherwise support your assertions concerning ACR. You're not giving the technical experts on this forum much to work with. Until you do, I wouldn't expect very useful responses.

Well a casual search of Google will show plenty of discussions about whether ACR adds NR even when turned off { not just to mFT cameras } . Apparently discussions are now evidence Joking aside there are several serious discussions on various forums and photography sites. I provided direct links to the RAW files I used what else do you think is needed , there will be all the data that is needed contained in the RAW files , post processed crops are of less value. Have a look at RAW files yourself in the respective RAW processors you favor. Looking at the comparisons between the RAW processors I have used does make it look like ACR adds some degree of color NR whether you have it turned on or off.

Unless you use lesser known far less common RAW convertors the mainstream options will honor the in-camera corrections by means of an instruction "header" op-code or whatever it is called.

Correct

Thus if you use any of the most popular { by a mile } RAW processors you will never see the non-corrected image so for most people who do not use the more unusual convertors , what exactly is the difference between a cooked file and what they will end up with ?

"Cooked" is a loaded term to begin with and you just compound things by trying to conflate software-based lens correction (which mFT lenses are deliberated designed to utilize) with raw-based noise reduction (which does not appear to be utilized with mFT cameras).

ACR,LR ,Aperture etc deliver a file which when opened has been processed according to the "recipe" "op-code" coming from the camera. That is cooked by any definition.The NR is an ancillary thought ,looking at various discussions it may be that ACR by default adds some NR to all RAW files { not just those from mFT} even when the slider is turned off.

[No further comment on the rest of your post, which is not relevant here or worthy of additional discussion.]

There will be no further responses from me to you either you have done your defensive duties regarding all things mFT and can hold your head high amongst your fellow guardians of mFT. So fare ye well good Knight

The_Suede wrote:

jim stirling wrote:

Thank you for your considered response.

...in which I forgot to include the thing I first thought about when I started writing it...

If you really want to know the "focal length" after the distortion correction, you can measure the included angle with the lens set at infinity focus.

Since it's a short lens, you could do this indoors, as long as you can shoot at F11 or something like that to get reasonable sharpness. The rest of the post is mostly grade school / high school trigonometry...

If you have a wall where you know the width, see how far you have to back off to include it in the image. Try to measure to the front pupil position, but you can probably assume that's in the front 5cm of the lens. The error if you get it wrong is quite small, as long as your aim target is more than a few meters wide. That gives you an equilateral triangle. Measure angles.

Search for the active width of your sensor (some 15-point-something mm?)

With that width as a base, draw/calculate the angles you measured to construct a small equilateral triangle.

The height of that triangle is your "focal length".

(no, it isn't, but it gives the number you're looking for at least...)

Yes, using that method the post distortion correction image appears to give you the focal length stated on the lens at least on the  Panasonic 12-35mm F/2.8 & 14-45mm. It also looks as has been mentioned that the correction instructions included with the RAW file when read in ACR etc may be being overly conservative and there may well be room to sneak a wider AOV with your own external adjustments