View Full Version : Removing rainbow artifacts
Other than the smart smoother iq plugin, what other plugins would do the job of removing rainbow artifacts often present in Anime?
On non-related note, is there anyway to apply a particular filter only to a certain portion of the video and leave the rest intact? For example, in the second part of End of Evangelion, there is a very noisy scene, but the noise is suppose to be there, and if I apply something like 2d cleaner and smart smoother hq, the scene will lose its intended effect, so I would like the filter to skip over that scene. Is there anyway to do that short of splitting the video and encoding the parts separately then joining them?
Thank you,
illbill
28th May 2002, 21:39
Well for your second question, yes you can do that without splitting the file first. You can download the "conditional filter" where you can specify for which exact frames and strength you want your filter to be used. It's just a matter of setting up a series of if statements.
You can get the filter at http://wwwtcs.inf.tu-dresden.de/~dc1/conditional/
Hope that helps
-illbill
Thank you, that's exactly what I needed. :-)
OUTPinged_
4th June 2002, 11:22
shu, no freeware tool is capable of removing rainbow effect from video.
Professional video editors can remove it but they use per-scene approach. Basically, there is no filter that will automatically remove/reduce rainbow artifacts from video.
There is an artice on why does it occur, but it is in russian :-)
RadicalEd
4th June 2002, 18:53
Actually (sorry to overrule your reply outpinged ^^; ) you can use SmartSmootherIQ (like you said) or check this thread;
http://forum.doom9.org/showthread.php?s=&threadid=25909
Since the rainbows are in the chroma plane this should be just what you're looking for as you can apply any sort of filter to chroma only.
OUTPinged_
5th June 2002, 11:38
radicaled, did that method prove working on these artifacts?
[update] Just tried it and it doesnt work.
You cant remove that rainbow only by analyzing chroma, you need to take into acount luma channel too, because it does occur when part of high frequency luma channel is mixed with liminance values of these pixels.
http://engel.otaku.ru/lab/dvd/dvdq_3.html <---here is a nice picture on how does it look.
I've had no success with what was mentioned in the thread, but it's a fact that Smart Smoother IQ does work, and all it says it does it to smooth the IQ plane. But however it does that is extremely slow (more than doubling my encode time). Has anybody ever read the code (I did, but I don't know a bit about color spaces or smoothing algorithms) and figure out or have a clue as to why it is so incredibly slow?
another way to tackle this prolem, will be reducing the rainbow artifacts during playback and not at encoding time.
something that works for me: ffdshow: on the 'resize' panel there's a 'chroma gaussian blur', just apply a 'generous' ammount of tat filter, and you'll probably loose these raibow effects.
cheers
avi
sh0dan
9th June 2002, 18:52
Originally posted by shu
Has anybody ever read the code (I did, but I don't know a bit about color spaces or smoothing algorithms) and figure out or have a clue as to why it is so incredibly slow?
It's slow for a number of reasons:
It does:
RGB -> YIQ
Smooth Q - plane
Smooth I - plane
YIQ -> RGB
So in essense it does four passes over every pixel - that has to be approximately multiplied by the diameter squared, to get the number of pixels tested.
It should be possible to get a >200% speedup by doing both smooths ain the same pass and replacing the convertions by MMX-code.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.