A distortion of the truth? Here’s why we’re not against software lens corrections: Digital Photography Review

The use of mathematical elements in a lens’s design can keep size down and still deliver excellent results. Photo: Richard Butler

One of the less obvious changes wrought by the move to mirrorless is that modern lenses don’t need to project an undistorted image. Now that there’s a processor between the lens and our preview of what the camera will shoot, it’s possible to apply digital corrections even as we’re lining up the camera.

Most of the major lens makers have embraced this possibility to create lenses that include mathematical correction, rather than solely optical designs. The first thing to recognize is that distortion correction isn’t being used to tidy up the results of a badly designed lens: modern lenses are designed with mathematical correction as one of their fundamental elements, with the rest of the optical formula planned around that.

Although it’s been commonplace for over a decade, this approach remains controversial. It can seem contrary to the belief that lens design is a case of striving for optical perfection, and some of the output can look pretty off-putting if you circumvent the corrections. Every time a lens designed for correction is launched, some website or YouTube channel will show an uncorrected sample image and say that the lens is somehow deficient.

Now that distortion correction is so widespread, we thought we’d take another look at why it’s done, how it’s changing lens design, and why we don’t think it’s appropriate to show sample images with these corrections omitted.

A key thing to understand is that distortion (along with lateral chromatic aberrations) is the aberration that’s most amenable to correcting mathematically. Whereas other aberrations tend to combine or spread the information from the scene in a way that can’t be undone, geometric distortion is essentially just a case of deforming the correct information into the wrong place. Geometric distortion only conflates information if you have too few pixels, such that adjacent information arrives at the same pixel and can’t be separated again.

Panasonic 14-28mm F4-5.6 @ 14mm | ISO 500 | 1/50 | F8 Photo: Barney Britton
Standard output	With corrections removed

But, while it looks visually disconcerting to see significant warping in an uncorrected image, it’s possible to re-map the data to the correct location with very little loss. Interestingly, whereas high pixels counts are often thought of as being more demanding on lenses (because they let you examine any aberration or imperfection in greater detail), they can make distortion corrections increasingly effective and accurate.

As one major manufacturer points out, it may well be that an all-optical design would produce better final results, but that requires cost, size and weight to be no object. And in any product, even at the high end, that’s rarely true. In the past year or so we’ve seen lenses that include mathematical elements produce excellent results from small lenses, and we’ve seen the release of lenses of a type that no one has previously managed to produce.

There’s some softness at the extreme lower left corner of the above image, but is it worse than you’d reasonably expect from a 14mm focal length (114° diagonal) image captured using a variable aperture zoom on a 47MP body?

Another important thing to recognize is that there are downsides to trying to correct everything optically. Optical engineers from Sigma and other manufacturers have confirmed that trying to perfectly correct geometric distortion with glass can lead to more complex designs with more elements, and the addition of these elements can then generate other aberrations, putting the efforts to control different aberrations in tension with one another. Using software to correct the aberration that’s mathematically correctable relieves some of this tension, allowing smaller, lighter, simpler optics with better optical correction of other aberrations.

Some of the correction examples we’ve seen appear really extreme, and are likely to immediately make you worry that the final image will need to be radically cropped, remapped and re-sized to be usable. Canon RF 16mm F2.8 | ISO 100 | 1/60 sec | F2.8 Photo: Chris Niccolls

Compatibility is a legitimate concern, of course. Some manufacturers are better than others at relaying their correction profiles to Raw processing software, and not all Raw processing software allows the manufacturer’s profiles to be applied. Having your software not be fully compatible with your new lens can be awkward if you have a well-established workflow. And, of course, it makes adapting lenses between systems more complicated, too.

However, none of this justifies giving excessive prominence to uncorrected images. In our opinion it doesn’t make any more sense to circumvent the digital element of a lens’s design than it would to decide we didn’t approve of aspheric glass and show the results for lenses with all those elements taken out.

The corners of the 16mm F2.8 look pretty dreadful, even when corrected. And yet even these improve to a pretty decent degree when you stop down, suggesting that the haziness doesn’t primarily stem from the amount of correction being done.
F2.8 upper left corner	F5.6 upper left corner

This doesn’t give anyone a free pass, though. If the digital correction of distortion lowers the image quality, we’ll show it and tell you about it in our coverage. But if it doesn’t, then we don’t believe it’s sensible to stoke concern over image quality impact a reader might instinctively assume will occur when correcting and cropping a distorted image. We believe the proof of the pudding is in the eating, not the specifics of how it’s prepared, we’ll continue to focus on the impact on the final photos.

With thanks to the industry experts, including Kazuto Yamaki and the engineers at Sigma, who helped check my logic in this article.