alias resizing_artifacts -> aliasing

BUR #5750 has been approved by @evazion.

create alias resizing_artifacts -> aliasing

Before getting technical let me say this is a case of tags being unknowingly used in parallel, aliasing first addition was on post #701000 in 2010.

The overwhelming majority of resizing_artifacts were used to denote in fact aliasing. resizing artifacts is a poor name for the effect since instead of merely denoting a flaw it requires any potential tagger to consider its cause which isn't always straightforward or possible.

The elephant in the room being if there is any practical situation where a resizing artifact is not aliasing. Yet in the matter of digital images most artifacts (say moire) are also a special case of aliasing. Anyway the tags need a clearer definition and I'm open to your expertise otherwise.

Reply NNescio user 499293 BrokenEagle98

DanbooruBot

over 3 years ago

The bulk update request #5750 (forum #187765) has been approved by @evazion.

Reply

ZipFile

over 3 years ago

BUR #5898 has been approved by @evazion.

remove alias resizing_artifacts -> aliasing

Applying BUR #5750 was a mistake.

Resizing artifacts was a catch-all tag for artifacts that occur both when downscaling and upscaling. It also covered general artist fuckups like post #3491600 (upscaled background) or post #889897 (upscaled flowers on the foreground). On the other hand, aliasing sometimes is an artist's fuckup too: post #3652038 (aliased 3d insert?), post #1479403 (bucket fill with low threshold?), post #2317322 (1 bit alpha channel). So it's not the same as resizing artifacts.

From technical point of view, aliasing resizing artifacts to aliasing is an error too: aliasing does not occur when image is upscaled.

more on upscaling/downscaling

When upscaling, the main problem is how to fill missing gaps. For instance, when upscaling 500x500 to 1000x1000 you have 4x amount of pixels. Naïve approach would be to map each pixel to new grid (e.g. [123,456] corresponds to [246,912]) and fill missing gaps with value of the previous pixel, a.k.a. nearest neighbor interpolation (example post #2576840). Less naïve approach is taking an average instead, a.k.a bilinear interpolation (like post #3268072), which results in quite jagged but at the same time blurred lines. Etc.

When downscaling, the problem is opposite: what value to use when pixels overlap. Say, for 1000x1000 -> 500x500 downscale, there will be 4 original pixels that correspond to 1. Super naïve approach is to discard extra pixels, this is a direct way to achieve aliasing (essentially, making conflicting pixels an alias to the resulting one). There are also number of ways to determine the pixel value, from different kind of averaging to more complicated techniques.

Moire as a defect is special. It may occur even if anti-aliasing filter was used: post #2558106 (smooth lines and text, but horrible halftone). It is not a special case of aliasing, even though they're often observed together.

Special case of resizing artifacts are ringing artifacts, they apply both to up- and downscaling. The effect is very subtle for downscales (post #2204621, post #1893671) and often destroyed by jpeg artifacts. When upscaling, the effect is more prominent: post #4270252.

Upscalers like waifu2x has their own set of distinctive "artifacts", a trained eye could pick up easily.

The overwhelming majority of resizing_artifacts were used to denote in fact aliasing.

The overwhelming majority of resizing_artifacts were various quality upscales, and only very few of them had aliasing due to downscaling.

The elephant in the room

The elephant in the room is that not everyone is so autistic to understand what resizing artifacts are and how to tell them apart.

Anyway the tags need a clearer definition and I'm open to your expertise otherwise.

Narrowing down resizing_artifacts to artist fuckups and third-party edits (like post #497605) that mix resized and non-resized components within the image (with visible artifacting, ofc) might solve the issue.