Pixiq

Upscaling has been on my mind this last week, with three things coming together at the same time. First came the submission notes sent by a stock library I’m joining, Aurora Photos. Although I’ve dealt with them by proxy for a number of years, via another site, now I’ll be sending files directly. They reminded me that the minimum file size is 50 MB uncompressed. For most of the archive this is no great deal, even though everything shot with the Nikon D3 needs some attention. Going back through the generations of equipment, however, there’s more of a problem, and when we get to the Nikon D100 that I was using in Sudan, the up-res is a real issue.

A 50 MB uncompressed image file translates to a little over 5000 pixels on the longer side for a full-frame 3:2 photograph (5120 x 3414). The file from a D3 is 4272 pixels on the longer side. My Sony A900 is over the limit at 6048 pixels, but a D100 is — ouch — just 3016. Getting from there to a 50 MB file means a 170% increase. And, looking a little deeper into those old D100 files, there are noticeably more artifacts and noise than from modern sensors.

So, upscaling has been necessary from the beginning of digital photography, and there is software to handle it. Regular re-sizing in Photoshop is probably the most used, and Adobe engineers have put a lot of work into improving the interpolation — both in Photoshop and in Camera Raw. This is now so good that gone are the days of Step Interpolation, which worked by reaching the desired up-res by doing a 10% incremental increase in resolution several times, one after the other. There is also dedicated software, and several years ago we tested what was on offer, and settled for what was then called S-Spline and is now PhotoZoom. The differences between the major players — Photoshop, S-Spline and Genuine Fractals — were not great, but on balance, for me, S-Spline came out best. Making a comparative test is not at all straightforward, because of the many variables, and because ultimately, there is no objective measure of success. Surely there is, you say. Well, as with noise reduction and sharpening, both of which these days come into up-res algorithms, there is always a trade-off between different qualities. The smoother you make the interpolation, the more risk there is of it looking ‘plasticky’ and unrealistic. Already, you see, there are some very subjective judgments creeping in here.

The second thing that happened last week was that my assistant and I went shopping for some large prints. The Epson 4800 that we have in the studio is fine up to a point, but I have a new show coming up on my Tea Horse Road book project, and for the launch we want some 2-meter prints. We discussed this with the print-shop that we finally settled on (Flash Photodigital in London). The owner, Howard Woodruff, uses Genuine Fractals, and said. “just give me the files and I’ll up-res them here”. That seemed OK to me, even though I don’t use GF myself. As I said, the last time I looked, there was hardly any difference. The size increase, by the way, from a 6048 x 4032 pixel Sony A900 file to make a print measuring 100 inches on the longer side (254 cm), is 500%, so we’re talking about the serious likelihood of artifacts.

But then the third thing that happened was PhotoZoom just launched a new version a few days ago — PhotoZoom Pro 4 — and they claim a new up-res algorithm that is better (well, they would, wouldn’t they?) but which also deals with aggressive JPEG artifacting and with noise. Now, this caught my eye, because if true it means being able to release for use images that were technically a bit iffy. I did a test (see the caveat above), taking a JPEG from a Leica D-Lux 4. I like my little Leica, but it’s not the world’s best for noise at night-time. And for the purposes of testing, I worked on a JPEG that had some noticeable compression artifacting, and gave it a significant, 400% up-res.

These are the results. The interesting new part in PhotoZoom is the Reduce JPEG Artifacts choice — a list of different strengths, from Light to Extreme, with the possibility of adjustment through what PhotoZoom call a Fine-tuning panel. For comparison (who said ‘comparisons are odious’?) I ran the image through Photoshop Bicubic. I also tried the latest Genuine Fractals, renamed Perfect Resize 7, but as the results were almost identical to those from Photoshop, have left them out from here.

...and the enlarged detail of these two...

There’s a trade-off, certainly, between artifact reduction and smoothness. The result here is simply my preference. But I have to say I’m impressed with the result.