12

I like to collect cool pics and use them as wallpapers or for other things. Often, artists publish only low-res versions, probably for fear of theft.

Example:

Gabriel Pulecio's BIRDS

Now, if I want to use that as a wallpaper, I'd have to upscale it, and obviously that'd make it look blurry because of the bicubic interpolation. I realize there's no real way to get a high-res version from a low-res pic, because the information is simply not there.

That said, I'm wondering if heuristics have been developed for upscaling with less apparent loss of quality. Those would probably be optimized for specific image types. For photorealistic pictures, for cartoons with large flat areas, for pixel art...

One algorithm I'm aware of is Seam Carving. It works for some kinds of pics, especially ones with a plain, undetailed or uninteresting background, and a subject that strongly stands out. But it's far from being general-purpose. Applying it to the above pic produces this. It looks quite sharp, but the proportions are horribly distorted because the algorithm is not designed for this kind of pic.

Another is Pixel art scaling algorithms. Those are completely unfit for anything other than actual pixel art that's pixelized to begin with. For example, I tried the scale2x windows binary on my pic, but its output was nearly indistinguishable from nearest-neighbour scaling because the algorithm didn't detect any isolated pixely fragments to work from.

Something else I tried was: I enlarged the image in Photoshop with bicubic interpolation, then I applied unsharp mask. The result looks pretty bad. The red blotch is actually resized reasonably well, but the dove is far from it.

What I'm looking for is some app that makes a best-effort attempt at upscaling any input image while minimizing blurriness. If you know of any, I'll be thankful. Note that the subjective prettiness and sharpness of the result is what matters... the result doesn't need to be completely faithful to the original small image.

2 Answers 2

7

Generally, fractal-based algorithms will turn up the best results. One example (which gets rave reviews, though I've not used it) is Perfect Resize. That said, as you mention, the data just isn't there -- especially by the time you account for (typically) the JPEG compression commonly found on web images. I think in a medium-sized image like the one you describe, you'll probably get decent results, but in smaller (such as thumbnail, or Flickr "small" size) images, nothing will get them to upscale to monitor size clearly.

3
3

I know this is question is pretty old, but I think Alien Skin Blow Up does a pretty impressive job. Here is your image scaled up 4x with it: http://www.picsarus.com/K7GUHI.jpg

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .