By: Erik Vlietinck
Tue, August 16, 2005 - 7:25:00
Pros: 0

Score:

FixerLabs is a British developer of photograph fixing applications. They developed the FocusFixer application which is capable of correcting lens blur for a large number of cameras in astonishingly accurate ways. SizeFixer is the first version of their upsizing program.

SizeFixer belongs to the same category of applications as Genuine Fractals. SizeFixer XL offers two features: it upscales an image and it sharpens that image using a highly sophisticated algorithm. The sharpening is done with the same technology as used in the FocusFixer application; it’s called super-resolution.

SizeFixer XL is capable of upscaling an image limited only by computer hardware. It uses EXIF data attached to digital images to build a model of the optics and the camera. This allows SizeFixer to reverse the softness in the original image as it upsizes.

Without EXIF data, SizeFixer will still work and upsize images with good results, but the amazement only comes with EXIF data present. SizeFixer should be applied before any other processing to an image, including the camera’s own sharpening algorithm.

SizeFixer works best on RAW image data, although I could get very good results from TIFF and JPEG as well. According to FixerLabs’ Tim Atherton, SizeFixer XL is a sort of supercomputer application for the Mac. It makes an initial guess at the high-resolution image, blurs that guess and downsizes it to compare with the original image. Any error between this down-sized guess and the original is used to improve the next guess.

SizeFixer may loop up to 100 times, improving its high-resolution guess each time around the loop. The Super-Resolution engine spends 98% of its processing time performing Fourier-transforms. The technology is very new, and according to Atherton, has previously only been used by NASA, the military and other government organisations.

If EXIF data is absent, a Fourier/Lanczos interpolator is used and can be applied by the built-in unsharp masking and edge enhancer. The Super-Resolution processing may take an hour or more to produce a result.

I tested SizeFixer XL with a variety of images, including photographs taken with my HP Photosmart 945, a Kodak P880 and a large number of images scanned or already on my disk from years back.

The program crashed on trying to use Super-Resolution with the HP image data. I sent the crash logs over to Tim Atherton, who came back to me within hours explaining that the EXIF-data from the camera isn’t 100% accurate. He pointed out to me the program wasn’t supposed to crash, and they were looking into why it did.

On the SizeFixer CD, there are two large photographs. I used each of those with success. They were RAW images, and Super-Resolution worked as advertised, complete with the lengthy processing times once I started upscaling the image to well over 500%.

On scanned images, the Super-Resolution feature would not work, but the USM and Edge enhancer did. Much to my surprise --after all, Photoshop’s unsharp masking filter is not a toy either-- the program’s USM and edge enhancer are exceptionally good; much better than Photoshop’s USM and without all the settings that make sharpening so utterly complicated.

Super-Resolution is what really makes this application unique. I tried it out on two images, one shot with a Nikon D100 and one with a Canon 10D. The first was a studio photo of a lady in night gown with a golden necklace holding an inscription. The second one was the pine cone image used on FixerLabs’ own web site.

I upscaled the photo of the lady in night dress and cropped the image to see if I would be able to read what the inscription read. With Super-Resolution I was capable of reading the characters that are visible on the photo. As a side-effect I was also capable of seeing every detail of the structure of that part of the subject’s dress.

The pine cone image is a good example of how the Super-Resolution algorithm is capable of making the upscaled photo sharper and just as detailed as the original. On the photo there is a water drop visible. On the original, this drop is more or less blurred. On the SizeFixer result at A1 enlargement, the water drop is literally crystal clear.

You might be excused to think that the rest of the photo --the cone itself-- as a result would be over-sharpened with those nasty halos we all know. In reality, the cone was sharpened but just enough to reveal all the little details that you would see if you looked at it yourself.

The workflow is deceivably simple: just select your output size in absolute or relative numbers, and the resolution setting for your output device. Then crop the image and go to the next tab. The Super-Resolution tab enables you to choose the balance between quality and performance and two quality settings (the sliders).

The same ease-of-use applies to USM and edge enhancement. While you’re working with SizeFixer you get the feeling that this can’t be all there’s to it, but it is. The program does all the rest. “The rest” can take a while. Even without Super-Resolution, upscaling an A4 image to an A1 took some 9 minutes to process and another 10 minutes before Photoshop had finally finished processing it for printing. The printing process itself took another 10 minutes.

The results, however, blow you away. In the below photograph that I took from the paper output, you can see how sharp the blown up image appears. It may not be clear from the photo that I took, but on the upscaled version you can actually see a few people standing behind their windows.

Conclusion

SizeFixer XL is an unbelievably powerful upsizing application. It made me think of those American spy movies where the hero digitally blows up a photograph to such proportions every detail becomes visible and the crook can finally be caught.

In the manual it says the better your camera is, the better SizeFixer’s results will be. I can imagine that very well. If your lens is of a very high quality, and your computer can handle it, I don’t believe there’s much that SizeFixer can’t reveal by blowing up the image based on what’s there in the digital data that we nowadays call a photograph.

Click on the image to see the larger photograph.