New Method For Depixelizing Pixels Disclosed
May 25, 2011 6:00 PM | Matthew Hawkins
Means of upscaling 8/16-bit visuals are hardly something new; anyone familiar with MAME knows already that it's able to smooth out those rough edges via filters and the such. Which many choose not to employ, especially these days. Those who wish to partake in the classics want the closest to the original package as possible.
Hence why the following will both confuse and possibly irk a good deal of those aforementioned folks: researchers have developed a totally brand new and supposedly sophisticated algorithm that reconstructs pixel-based images into vector graphics. What you see above is "a complex blend of pixel analysis and spline curves". How this is any different from what MAME already does is not exactly clear, plus the fact that the author of the article chooses to refer to the dolphin from Super Mario Word that's used as a test subject as 8-bit will no doubt rub people the wrong way further (it's 16-bits, duh).
The real question is obvious: what's the point? Especially these days, when appreciation for pixel art is at an all-time high. Yesteryear's visual style no longer has the stink of inferiority across the board; aside from being embraced as something culturally significant among diehard gamers and art aficionados, among mainstream gamers and the general public, it's something nostalgic and simply "cool" again (aka marketable).
Remember when the first screens of the XBLA version of Guardian Heroes were revealed? Everyone got up in arms over the smoothed out visuals, and the pitchforks were only lowered when it was revealed that the original graphical mode was in tact. Then again, advances in technology doesn't necessarily have to make people happy per se, and it might have some other application, outside of gaming. Perhaps to enhance low-resolution video feeds, as the first commentator in the source states. Who also says: "To hell with video games."
[via Extreme Tech]
Categories:
3 Comments
It's different from anti-aliasing, which basically means blending the edge pixels with their neighbors. This is actually looking at image and constructing a new 'best guess' based on it - picking where the outline is, how the curves should smooth out, etc.
In this particular application it's not amazingly useful, but these are the kinds of techniques that graphics programs (like Photoshop) and graphics hardware need to use in order to keep file sizes and memory costs small and output pretty. For instance, 3D programs do the same kind of thing - you create a model with really flat, discrete polygons, and then when it's rendered, the whole shape appears smoothed out and curved.
Vin St. John | May 26, 2011 8:08 AM
I think it is quite useful.
One, for make weird art pieces from old sprites.
And two, to make another filter for emulators (maybe the algorithm is more quick than the Hqx one, who is pretty intensive).
You know if there is this kinds of filters in emulators it is because people use them. I use them, I like them. I usually select one preferred for each game, so each game shows pretty pretty. Castlevania or Zelda II for NES with Hqx are amazing.
Ruber Eaglenest | May 26, 2011 11:03 AM
Unfortunately, the linked article mentions and then pretty much brushes aside one of the big issues with any of these depixeling routines, that they don't always work.
It even explains why they don't always work. Every pixel is important, but these algorithms don't know whether something is supposed to be smooth or sharp, and guessing wrong can range from acceptable to silly.
When you apply one of these filters to a game working in real time, you get both the good and the bad. You might get nice backgrounds and some characters that look great, but you also get a weird looking font and maybe projectiles look wimpy. Or maybe some characters look good while others look bad.
Billy Bissette | May 26, 2011 9:46 PM