A distorted view? In-camera distortion correction
Richard Butler | Technology | Published Sep 2, 2011
In-camera distortion correction: Can software ever take the place of optical excellence?
Richard Butler looks at the trend towards cameras correcting lens distortion and what it means for photographers.
What's more important, the final image or how it's achieved? Over the past year or so, we've encountered an increasing number of cameras with lenses that capture significantly distorted images which are then corrected using in-camera software. It's a process that has divided the forums, with some arguing passionately that it's cheating and that the lenses aren't really as good as they appear, while others maintain that it's the final results that matter, rather than how they're achieved.
In the film era, distortion correction had to be designed into the imaging lens, and this was (and remains) true for digital SLRs too (the need for the captured image to match what the photographer saw through the viewfinder limits what the camera can do to the shot once taken).
The advent of live view digital cameras (fixed and interchangeable lens) not only allows software correction of distortion (something impossible in an analog camera), but crucially also lets you see the result of whatever distortion corrections are being applied at the point of shooting, so the preview image matches the captured image.
There's no reason to believe this is a particularly new approach for camera makers (it would help explain the sudden trend for including 28mm equivalent wide-angle zooms in compact cameras), but its appearance in enthusiast cameras such as the Panasonic DMC-LX3, which output RAW data, means it has become potentially visible to the end user.
With the arrival of Micro Four Thirds, we now have an interchangeable lens system with many lenses that require software correction of distortion. So what are the manufacturers playing at? Why are they releasing lenses with distortion, what effect does it have, and should dpreview review the distorted or the corrected output? We spoke to a series of leading camera and lens makers, as well as the creators of leading RAW converters to find out just what's going on.
So what do we mean by distortion anyway?
First of all, it's important to understand what we mean by distortion here. Quite simply, we're talking about the lens rendering lines which should be straight as curves. This is is known as rectilinear distortion, and comes in two basic types: barrel distortion, where lines bow outwards away from the center (making a rectangle resemble a barrel), and pincushion distortion, which is the opposite.
Why have manufacturers left this distortion in their lenses?
The simplest answer to this question is that they can. On film, if you captured a distorted image on your negative, it would be very difficult to subsequently correct it. As more powerful in-camera processors have become less expensive, distortion can be corrected and, because many cameras offer only electronic displays, the user can frame and compose their shots based on a pre-corrected preview.
The preview on the back of the Canon S90 shows the same corrected image as will appear in the final image, with a closer relationship between the framing of the preview and final image than you'll get with most DSLR's, most models of which tend not to have viewfinders offering 100% coverage of the scene.
Olympus, a founding member of the Micro Four Thirds system, explains the possibilities this creates: ‘this lens distortion correction by software enables us to greatly reduce the length and volume of a lens, and gives manufacturers flexibility of small and light weight designs.'
'By using the software correction, front lens groups with strong optical power are not necessary, and the size of the front optical element can be reduced. As the total length of the lens can shorten, it is easier to design light weight and compact size lenses.'
Canon, whose PowerShot S90 also applies correction to its output, makes a very similar point: ‘In designing any optical system we need to consider multiple aspects including cost, target market, product size, manufacturing complexity / efficiency and sensor. It gives more flexibility to the design of the whole system if software correction is considered.'
Kazuto Yamaki, Chief Operating Officer for lens maker Sigma explains how this can work in more detail: 'compromise in distortion and chromatic aberration is a very effective in making lenses more compact and reducing cost. In general, compromising the distortion at the wide-angle end affects the compactness of the lens and compromising the CA influences cost reduction.' It's not without its drawbacks, he says: 'it's also true that doing this deteriorates the image quality if we correct them with image processing. It may not be very obvious if you look at the images in small format, but it becomes recognizable when enlarged. In the case of users who are very, very insistent on top image quality, they may not be 100% satisfied with such images.' As a result, no Sigma lenses currently require software correction.
The extent to which software correction is used depends on the purpose of the lens, he says: 'If the very best image quality is the desired goal then we might not be able to use it but if compactness and reasonable price is considered a priority over the image quality (for those who want such lenses), we can use that methodology.'
All the manufacturers we spoke to make clear that the basic lens design process isn't fundamentally changed when developing a lens that will include a degree of software correction – they aren't using optically-uncorrected, fish-eye lenses or anything so extreme. However, Olympus sheds a little light on the process: ‘Generally an optical correction of barrel distortion tends to cause a moustache distortion,' which is very difficult to correct for in post-processing. ‘We consider that we can improve the final image quality with a balance of optical distortion corrections and software correction,' the company says: ‘rather than with only with an optical design solution.'
It's a question of balance between the image quality and size/price, explains Sigma's Mr Yamaki: 'we can compromise somewhere between the two - correct the aberrations nicely in the lens design and also correct them non-aggressively in the image processing. It's not the matter of choice between the two options.'
What's being corrected?
From our testing, we've seen different manufacturers take different approaches and to different degrees. Panasonic's Micro Four Thirds lenses contain information about both geometric distortion and chromatic aberration (color fringing), Olympus's only contain geometric distortion information. Panasonic's camera bodies correct whatever they receive information about from the lens, and also pass that information on to their RAW files, while Olympus' only correct and retain distortion data. Samsung tells us its NX series doesn't yet do any correction but has the capability to do so if needed in future lenses.
What's the cost?
Understandably, many people are wary of the idea of what they perceive as 'optically imperfect' lenses and the image manipulation that consequently has to go on. However, while it is easy to make assumptions and to criticize as a point of principle, it's harder to prove any great damage is being done.
Here we've taken the same shot using the Olympus 14-42mm kit zoom on an E-620 DSLR and using the corresponding Micro Four Thirds lens on an Olympus E-PL1. The DSLR version of the lens produces 1.38% distortion at this setting, according to our lens test, while the Micro Four Thirds version natively produces 4.5% distortion (one of the highest levels of distortion we've measured), and corrects it to around 1.8%.
Our copy of the DSLR lens seems to have developed a degree of asymmetry since we last checked it, but is likely to still likely to be representative of kit lens performance in the real world. They are, in any case, the closest we could find to comparable systems, one of which relies just on optical correction, the other of which combines optical and digital correction.
|E-620 with Zuiko Digital 14-42mm||E-PL1 with m.Zuiko Digital 14-42mm|
Overall, the best corners of the DSLR lens that makes use of a purely optically-corrected design have an edge over the Micro Four Thirds lens that combines both optical and digital correction. But as you can hopefully see from these crops, it's pretty slight - you have to look at the pictures very closely to notice any difference.
What if I don't want to use the manufacturer's RAW converter?
In all the instances we've encountered, the cameras are recording un-corrected data in their RAW files and including information about lens distortion characteristics as additional information (metadata) alongside the underlying capture data in the RAW file. However, Panasonic says that the Micro Four Thirds standard allows for both approaches: ‘it depends on manufacturers' policy whether corrected image or native image would be recorded.' So this may not always be the case in the future.
Canon, Olympus and Panasonic all make clear that they provide their own RAW conversion software with their cameras that automatically corrects images. However, although metadata describing the distortion is included in the RAW file, it is up to individual software makers how much of this information they use when making conversions; and Olympus says it does not disclose its RAW file format to third-party software providers.
In co-operation with camera makers, Adobe has developed both its Camera Raw (ACR) converter and the open DNG format to recognize and incorporate distortion correction parameters. This means you get the same geometric distortion corrections for supported cameras as in the manufacturer's software.
Adobe Camera Raw
DXO Optics Pro
DXO Labs' DXO Optics Pro software, meanwhile, is based around lens profiles developed by the company based on camera testing, rather than data supplied by manufacturers. At the time of writing this article, neither the Panasonic G1 nor Olympus E-PL1 were supported by the software, however.
Meanwhile, Phase One says its Capture One converter provides ‘native lens correction for certain cameras and lenses. The support is dependent on the manufacturer's willingness to provide information on lens errors and correction requirements.' Capture One Pro also provides a lens correction tool allowing the correction of geometric distortion, fall-off and CA based on individual lenses: ‘there is a substantial difference in using a generic model and using a model where the measurements/correction take place with basis in a specific lens.'
|Out-of-camera JPEG||Capture One Pro 5.1||dcraw|
Like DxO Optics, Capture One Pro applies its own levels of lens correction but again doesn't
support all the cameras we tested. Interestingly it applied automatic correction to the
Panasonic DMC-G1 images but not the Canon S90's.
Overall, the result is that, whether you're using the manufacturer's software or the most popular third-party conversion software, you'll get the same undistorted image that you composed on the cameras' screens or viewfinders and that you'd get from the camera's JPEGs. But some other raw converters won't provide properly corrected output, so if you choose to use one of these, you'll have to fix the distortion yourself.
Isn't it cheating, though?
Every time somebody does anything with a digital camera that couldn't be done in a film darkroom, people have called 'foul' and make accusations of cheating. However, 20 years after the appearance of Photoshop, it's safe to assume that a degree of post-shoot 'retouching' is the norm, rather than a sneaky exception (And remember that plenty of secrets could hide in the darkrooms of skilled practitioners). At which point, there's a chance that one person's cheating might turn out to be progress for the majority.
After all, this isn't the first instance of automated, non-optional image correction that has no darkroom analogue. Most current Nikon DSLRs assess their images for lateral chromatic aberration and correct it in their JPEGs. This could just as easily be seen as cheating, since it results in a final image that hides some imperfections that optical design alone hadn't eliminated from lenses. The situation is slightly different in RAW, since no correction data appears to be included with the file but, given that Nikon's own software applies exactly the same corrections and that many third-party converters can also apply similar lens corrections (either automatically or controlled by the user), the distinction is subtle.
The only fundamental difference is that most Nikon lenses pre-date the application of these corrections, so can't have been designed to take them into account. However, only a handful of people within the company will ever know whether the correction is considered when developing new optical formulae. After all, if fractionally under-correcting CA allowed a slightly simpler, lighter, sharper or less distorting designs, then wouldn't such a design be preferable, if the cameras were going to remove that additional CA anyway?
What about reviews on dpreview?
This brings us back to our original question: ‘What's more important, the final image or how it's achieved?' because it's one we have to answer every time we review a camera or lens. It's our job to give readers a good understanding of the product they might be thinking about buying, but does it make sense to criticize it for something that most people will never be need to be aware of?
Lens reviewer Andy Westlake explains the approach the site takes: ‘In our opinion, what matters most is the final image quality that the end user is most likely to see - which means what they'll get from JPEGs, or raw files developed using either the software supplied with the camera, or the most popular raw converters such as Adobe Camera Raw.'
This informs the attitude taken in reviews, he explains: 'we'll be basing our conclusions on the results you'll get when corrections have been applied - and therefore measuring any loss of sharpness that results from 'stretching' at the corners. But we also think it's important to show the user exactly what corrections are being used, so we'll also show how much distortion and chromatic aberration is present in an uncorrected image.'
Finally, he says, it's important to keep things in perspective: 'There seems to be a tendency to assess the performance of a lens from its sharpness alone, but we think there's more to it than that, and it's the balance between sharpness and image defects such as distortion and chromatic aberration that really counts when assessing a lens. For example some conventional SLR zoom lenses exhibit very strong barrel distortion at their wide ends which is complex in character and difficult to correct in software, and this can spoil an image no matter what size its printed or displayed. The use of software correction simply gives manufacturers another tool they can use to address such problems and deliver better-looking images with less distortion and colour fringing - and as long as it doesn't have too much of an impact on sharpness, we believe this is surely a good thing'.