New Method for Vectorizing Pixel Art

Hot Topics:

By Sean Michael Ragan
Posted 2012/06/04 @ 1:00 pm
Category Art & Design

While algorithms for turning raster images into vector images are nothing new, the usual suspects perform terribly when it comes to sprite-sized pixel-art type images where each pixel matters a lot. Enter this cool project that Johannes Kopf and Dani Lischinski presented at Siggraph 2011. Their paper, which nicely explains the nitty gritty, is freely available (PDF), and there’s also an online gallery where you can see how their method stacks up against a bunch of other algorithms when it comes to scaling up some classic sprites. I especially like the new streamlined Space Invader shapes. I always wondered what they really looked like. [via O'Reilly Radar]

Depixelizing Pixel Art

More:
Image deblurring using inertial measurement sensors

In the Maker Shed

9 Responses to New Method for Vectorizing Pixel Art

fire115 on June 4th, 2012 at 1:41 pm said:

Amazing!

Reply ↓
MP on June 4th, 2012 at 2:11 pm said:

This was from Siggraph 2011. Has anything changed in the last year?

Reply ↓
DD on June 4th, 2012 at 10:16 pm said:

did anyone notice that for Invader 1 on :
http://research.microsoft.com/en-us/um/people/kopf/pixelart/supplementary/multi_comparison.html
Vector magic turns a Space invader into an Enterprise crew member ?

Reply ↓
MarkkuA on June 4th, 2012 at 11:34 pm said:

When will we have this as a plugin for illustrator?

Reply ↓
Idgarad on June 5th, 2012 at 6:05 am said:

You do realize this is just someone trying to rip off the Sai and Eagle depix algos that have been used in NES\SNES emulators for over 15 years right? Check out ZSNES as a quick example.

Reply ↓
- Sean Ragan on June 5th, 2012 at 9:13 am said:
  
  No, it isn’t. The algorithms you refer to are image scaling algorithms; they output larger raster images that are smoother than those produced by naive nearest-neighbor upsampling. This algorithm outputs vector graphics. If you’ll actually click through to the gallery linked, above, you can do side-by-side comparisons and see the difference between this new method and the Sai and Eagle algorithms you mention explicitly. The new method is clearly superior, and works by a very different mathematical process.
  
  Reply ↓
rinf on June 5th, 2012 at 12:30 pm said:

Idea: Integrate this algorithm into an NES/SNES emulator and make Super Mario Bros. a vector graphic game. I really wonder what it might look like.

Reply ↓
- mod3 on June 6th, 2012 at 3:26 pm said:
  
  they thought of that and gave an example of how it would look
  http://research.microsoft.com/en-us/um/people/kopf/pixelart/supplementary/video_ours_4x_h264.mp4
  
  Reply ↓
johney on November 4th, 2012 at 2:29 am said:

Mind blowing,now if only also the colors will be dithered,then it will look much smoother with no flatty colors anymore,the reason why those sprites still does,nt look complete such as bowser missing some lines on the edges and in the contours such as his right teaf and sharp amps,was because when nintendo have converted their drawings into low_res sprites,some details were lost and cannot be recovered when vectorizing,so some manual intervention isrequired to correct this,
or they shut future enhance the algorithm wich may be able to follow the directions of colors inside and on the edges of these sprites and inserts thin lines in between each color and thick lines on the edges of those sprites to make it look much more complete,sure it will very complex to create such interpolator and that will require more horse power,but once done,those backgrounds and sprites will look more better then ever before!!!!

Reply ↓