Depixelizing Pixel Art | Hacker News Reader

Depixelizing Pixel Art

johanneskopf.de

74 pointsjhack20 comments

BTW, there's a video of some Mario game upscaled with this algorithm. Looks very modern! http://johanneskopf.de/publications/pixelart/supplementary/v... (3MB)

alt here: http://ifile.it/gj1tlk5/video_ours_4x_h264.mp4

Alternate alternate mirror here: http://www.filesonic.com/file/1061063604/video_ours_4x_h264.... had trouble with ifile.it at work)

"You may have reached this page due to a known problem with some UK ISPs" with absolutely no description of what that known problem that they know about actually is. Very bad UI decision there.

I'm pretty sure I can guess what the problem is (the IWF filter, no doubt) but if I were an ISP and got a pile of reports from users simply parrotting that message I'd block the site completely until they improved it. It is like me as a programmer getting "I gots an error message" from a user who then says "oh, I didn't read it all" when I ask what the error message actually was.

Awesome stuff. This is the link to the PDF:

  http://johanneskopf.de/publications/pixelart/paper/pixel.pdf

It takes a while to start, but I just downloaded it now...

Here's the document in image format: http://imgur.com/a/gRXPJ

Anybody happen to have a mirror of the PDF?

I've found only Google cache: http://webcache.googleusercontent.com/search?q=cache:15iKIgS...

Unfortunately it lacks images, which is critical in this case :-/

I managed to download it. Where could I mirror it? If I give out my dropbox public link, will I get lynched?

Try http://www.scribd.com/

It does download successfully if you wait long enough (~5 min).

http://www.scribd.com/doc/56137278/Depixelizing-Pixel-Art

If you put it on Scribd don't you have to declare that you have the right to do that? Don't you have to declare that you are the copyright holder, or that the work is free to copy?

I love the image caption regarding anti-aliased cases: A less successful case. Anti-aliased inputs are difficult to handle for our algorithm. “We are doomed...”

Further discussion in a duplicate submission a week later:

http://news.ycombinator.com/item?id=2601347

This is actually nothing new. Without meaning to criticize the author's efforts, the whole paper has a strong feel of being "reinventive". There are several algorithms for this in use since long (almost all of which were made specifically to improve video game console emulators) such as the early EPX algo from 1992, and the newer MAME algorithms and hqx.

http://en.wikipedia.org/wiki/Pixel_art_scaling_algorithms

They compare their results to hq4x, and arguably come out ahead.

Has the same algorithm been proposed before? (I take it that it hasn't. So it is indeed new ...)

Actually, it is in a completely different league.

The algorithms you linked to output bitmaps.

Kopf and Lischinsk's algorithm is a vectorizer, optimized for old school graphics. This means that its outputs is resolution-independent. Similar tools include Potrace[1] (open source, but for black and white images) and Vector Magic[2] (multicolor, but proprietary).

However, the use of Voronoï diagrams is AFAIK original, and their results are superior to hqNx at similar resolutions.

[1]http://potrace.sourceforge.net/

[2]http://vectormagic.com/home

you obviously did not get it..

Reply on news.ycombinator.com