Pixel Art Scaling Comparison Part II

I’ve been playing with a few other scaling algorithms in the 2dimagefilter project. They produce even more impressive results than the previous batch…

Part I showcased the Eagle, SuperEagle, SaI and SuperSaI algorithms. I will continue here with EPX/Scale, HQx, xBR and xBRZ.

EPX/Scale

The EPX algorithm was originally developed by Eric Johnston at LucasArts around 1992.

Nine years later, while working on the AdvanceMAME project, Andrea Mazzoleni developed the Scale algorithm, which produced exactly the same results as EPX, even though Andrea didn’t know of its existence. Great minds think alike, I guess.

Here you can see an image scaled using Scale, compared to the nearest-neighbor algorithm:

And here you can see the details:

Great results. I can see why so many emulators include it.

HQx

The HQx family of algorithms (HQ2x, HQ3x, HQ4x) was created by Maxim Stepin around 2003. They are all based on pattern recognition using a pregenerated lookup table.

Here you can see an image scaled using HQx, compared to nearest-neighbor:

And here you can see the details:

Excluding the blurriness of the roof pattern, it does a great job with this image.

xBR

The xBR algorithm was created by Hyllian in 2011. It is also based on pattern recognition, but uses a multi-stage set of interpolation rules.

Here you can see an image scaled using xBR, compared to nearest-neighbor:

And here you can see the details:

The results are much sharper than what HQx produced for this image. Extremely impressive.

xBRZ

The xBRZ algorithm was created by Zenju in 2012, as an enhanced and optimized version of xBR. It equals the performance of HQx, and on certain architectures it even surpasses it.

Here you can see an image scaled using xBRZ, compared to nearest-neighbor:

And here you can see the details:

I can’t tell it apart from xBR. Everything looks super sharp.

That’s all for now, I guess. If you enjoyed this, be sure to check out the paper Depixelizing Pixel Art by Johannes Kopf and Dani Lischinski. The supplementary material also contains a ton of interesting comparisons between algorithms.