What's a good rainbow remover?

[Mephesto]

I do not expect an optimal one because I don't see it being possible to eradicate selective chroma flickering without false positives, but this forum has proved me wrong many times when I had people who've done video restoring for a living insist otherwise.

I tried Bifrost, LUTDerainbow, Derainbow and SSIQ from Vdub. Bifrost and LUTD did nothing. Derainbow is a more impressive filter that greatly reduced rainbowing but dulled many vibrant colors of false-positive objects. SSIQ's strength was damn noticeable too, even more noticeable on regions where it's unwanted.

[poisondeathray]

It depends on what the rainbows are from, how severe, the distribution...etc...

You can also try chubbyrain2, tcomb with chroma settings (mode=1 plus other settings)

I don't see it being possible to eradicate selective chroma flickering without false positives

You might not be describing it properly. What you call "selective chroma flickering" might not be what rainbows , it might be something else - maybe a small sample would help

[Mephesto]

http://www.sendspace.com/file/zs0u65

I also forget to mention I tried mfrainbow but couldn't get filter to work.

I tried chubbyrain and it smeared the rainbows instead of removing them.

[poisondeathray]

Oh, this is your messed up Southpark source...

I tried chubbyrain and it smeared the rainbows instead of removing them.

Like all derainbow/dotcrawl filters, you should use it before any other filters (so while it's still interlaced, before srestore, before everything else - it should be the 1st filter after the source filter)

It's chubbyrain2, not chubbyrain ; some frames have residual artifacts so you can tweak the settings, stack something else , or use a mask with strong chroma filtering

[Mephesto]

Oh, this is your messed up Southpark source...

Psssht!

Like all derainbow/dotcrawl filters, you should use it before any other filters (so while it's still interlaced, before srestore, before everything else - it should be the 1st filter after the source filter)

If I do that it will screw up the interlacing and cause color ghosting later.

It's chubbyrain2, not chubbyrain ;

My bad. I first typed "tubbychubbyrain" and was ready to bust a cap when avisynth said:

some frames have residual artifacts so you can tweak the settings, stack something else , or use a mask with strong chroma filtering

I couldn't find any documentation for chubbyrain2. What are some things I can tweak? I noticed setting to -1 filters the strongest but even then it misses spots.

[poisondeathray]

haha nice error message

Like all derainbow/dotcrawl filters, you should use it before any other filters (so while it's still interlaced, before srestore, before everything else - it should be the 1st filter after the source filter)

If I do that it will screw up the interlacing and cause color ghosting later.

Your source is foobar to begin with, but the fact is all derainbow filters and dotcrawl filters work best before any other filter, while it's still in fields. You will find they almost don't work at all otherwise. Then you have to blur the shit out of it to get anything to work. You decide how you want to filter it and what settings to use . This is supposed to be a progressive source, but has gone through too many bizzare manipulations so maybe the rules don't apply anymore

I dont of any documentation for chubbyrain2, it's one of those quickly put together ones . You can play with the radius or threshold settings, or you can stack it with other filters

[sanlyn]

script of function chubbyrain2:

function ChubbyRain2(clip c, int "th", int "radius", bool "show", int "sft")
{
#based on Mug Funky's ChubbyRain

th = default(th,10)
radius = default(radius,10)
show = default(show,false)
sft = default (sft, 10)

u = c.utoy()
v = c.vtoy()

uc = u.mt_convolution(horizontal="1",vertical="1 -2 1",Y=3,U=0,V=0)
vc = v.mt_convolution(horizontal="1",vertical="1 -2 1",Y=3,U=0,V=0)

cc = c.mt_convolution(horizontal="1",vertical="1 2 1",Y=2,U=3,V=3).bifrost(interlaced=false).cnr2().temporalsoften(radius,0,sft,2,2)

rainbow=mt_lutxy(uc,vc,Yexpr=string("x y + "+string(th)+" > 256 0 ?")).pointresize(c.width,c.height).mt_expand(y=3,u=-128,v=-128)#.blur(1.5)

overlay(c,cc,mask=rainbow)

show==true? rainbow : last

}

I think it requires these plugins/script (I first used it a long time ago, but my old notes say these were required):

masktools v2
Bifrost
LimitedSharpenFaster
RemoveGrain v0.9

discussion (one of many): http://forum.doom9.org/archive/index.php/t-143438.html

Recent example of its use: https://forum.videohelp.com/threads/322848-Scripts-and-filters-for-an-Opera-%28sample-v...=1#post2000913

[poisondeathray]

The problem (especially with this source) is it has gone through too many conversions with analog, PAL, NTSC, and there is blending, so any autodetection mask will fail, either will miss areas, or overshoot areas . That's the same reason why the usual dotcrawl filters failed in your other thread, and damaging , heavy overkill filters were required to get rid of all the dotcrawl

You can see the mask used by chubbyrain by using show=true, and tweak it using mask tools , it will be difficult to isolate ONLY the rainbows and nothing else on this source

One way you can improve the results is use a heavy filtered version, with a lightly filtered version, with mt_merge , at least you can limit the damage that way

[Mephesto]

This is supposed to be a progressive source, but has gone through too many bizzare manipulations so maybe the rules don't apply anymore

There is no progressive source for the early seasons, not until they release Blu-ray versions. This is the best I can do for now.

I dont of any documentation for chubbyrain2, it's one of those quickly put together ones . You can play with the radius or threshold settings, or you can stack it with other filters

I did, and I'm lost. Increasing both to 20 strengthened the filtering very slightly. Decreasing to 1 strengthened it noticeably, and -1 strengthened it even more. Going above 20 nullifies the filter completely. Actually, going above 10 nullifies it.

Oh well, I've decided to use it selectively on scenes where rainbowing is particularly prevalent. I don't even know if it's that big of a deal TBH. My main concern is removing illegal entropy from the video, because we're aiming to match both DVD quality and the famous RM bitrates.

Chubbyrain2 is the best out of all the derainbowers I've seen so Mr. Mackey will be pleased to see the end of his little rainbow infestation problem in his hair.

Thanks 'ray.

@Sanlyn, the error message was a joke, chubbyrain2 worked fine.

[poisondeathray]

If you want an overkill script

a=MPEG2Source()

a
chubbyrain2(show=true)
msk=last

a
chubbyrain2(th=1, radius=15)
base=last

base
fft3dfilter(sigma=16, plane=3, bt=4, interlaced=true)
heavy=last

mt_merge(base, heavy, msk, luma=true)

You will notice it gets rid of almost all the rainbows, but also color edges start to get eroded. You view the mask and view how it overshoots many areas. You would have to modify the msk layer to get more accurate, there might be some novel ways, you can ask at doom9. Adjust the sigma value to adjust the "overkill" settings. You can try to sharpen the chroma at a later stage with mergechroma(awarpsharp2 (#somesettings))

You have to ask yourself what is the lesser of evils ? Can you live with some rainbows or would you rather have fewer rainbows and blurry color edges

I don't know of anyway to get a "perfect" mask automatically, because potentially anywhere 2 colors reside beside each other is a potential "rainbow"

[sanlyn]

@Sanlyn, the error message was a joke, chubbyrain2 worked fine.

Oh. I thought you couldn't find the little critter.

Methinks poisondeathray is generally correct. There is such a thing as overfiltering.

[Mephesto]

a=MPEG2Source()

a
chubbyrain2(show=true)
msk=last

a
chubbyrain2(th=1, radius=15)
base=last

base
fft3dfilter(sigma=16, plane=3, bt=4, interlaced=true)
heavy=last

mt_merge(base, heavy, msk, luma=true)

Can you expand on this? This is greek to me.

You will notice it gets rid of almost all the rainbows, but also color edges start to get eroded.

What more concerns me is the destruction of the colored interlace combs which causes color-ghosting upon deinterlacing later. I'm impressed with the filter however. The removal of the rainbows is absolute and erosion of edges is minimal. I still cannot allow this globally but I won't mind doing it on scenes with severe rainbowing.

You have to ask yourself what is the lesser of evils ? Can you live with some rainbows or would you rather have fewer rainbows and blurry color edges

I would rather have only Mackey be an acidhead and not everybody else hallucinating vibrant rainbows when they are forced to see him for counselling.

I permit rainbows only in the sky.

I don't know of anyway to get a "perfect" mask automatically, because potentially anywhere 2 colors reside beside each other is a potential "rainbow"

That's why I didn't expect any optimal filtration since my first post in this thread, but chubbyrain2 had more balls than all the other derainbowers combined and worked much better than I anticipated.

This is yet again one of those times when videohelp has proved me wrong. Thanks guys.