How can I perform an ABX test on different AVIsynth filters?

[hkl8324]

How can I perform a ABX test on different AVIsynth filters? (the filters doning the samililar things)

Or do we had a already done ABX test on those AVIsynth filters?
(By function, again)

People here always said XXX (Insert a filter name here) will do better than YYY (Insert another filter name here) for a particular job. By yet we dont have a objective prove...

[foxyshadis]

The point of avisynth is to perceptibly change the stream, it'd be like ABXing different settings of a graphic equalizer or flange. Rather silly. (Unless you're attempting to recreate someone else's effect, perhaps.) ABXing a codec is the only thing that would make sense.

Objective proof comes in two flavors: Encoded file size, and comparison to a normal. The normal test used is to add a random grain to the image and apply the filter, then compare the before and after. (If you cared to, you could abx that, for instance.) The problem is that artifacts are often more complex than gaussian noise and difficult to simulate.

None of this can really compare with simply looking at the output of several filters side by side and interleaved and picking the one that you feel does the best job for your particular combination of video and artifacts. People develop personal preferences of how they like the video to look and use favorite filters for that.

[Richard Berg]

ABX testing only tells you whether two clips are perceptually different. It can't tell you which is "better." For that you'll need a another methodology, and inevitably, more statistics. ANOVA is popular for testing audio codecs, for instance.