Heavy rainbows on video

[kenator18]

Hi everyone,

I'm having trouble removing rainbows on a music video, some help would be gladly accepted! I've tried about every single de-rainbow script out there... Here's my script and file down below:

MPEG2Source("F:\D2Vs\One In a Million.d2v")
bob(-0.2,0.6)
srestore(frate=23.976, omode=6)
crop(10,6,-10,-4)
Lanczos4resize(720,540)
Checkmate(24)
DeRainbow()
SmartSSIQ(strength=200, interlaced=false)

Thanks!

[themaster1]

Try this:

##Load filters:
LoadPlugin("C:\Program Files (x86)\AviSynth 2.6\plugins\KNLMeansCL-v1.1.1\KNLMeansCL.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth 2.6\plugins\TTempSmoothv094\TTempSmooth.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth 2.6\plugins\variableblur0.7_x86_x64\x86\variablebl ur.dll")
#####
MPEG2Source("C:\OIAM.d2v", cpu=0)
ConvertToYV16(interlaced=true) #
orig=last
ev=orig.assumetff().separatefields().selecteven()
od=orig.assumetff().separatefields().selectodd()
ev
ue_chroma = UToY(ev).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
ve_chroma = VToY(ev).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
YToUV(ue_chroma, ve_chroma)
MergeLuma(ev)
ev_filtered=last
od
uo_chroma = UToY(od).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
vo_chroma = VToY(od).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
YToUV(uo_chroma, vo_chroma)
MergeLuma(od)
od_filtered=last
interleave(ev_filtered,od_filtered)
assumefieldbased().assumetff().weave()

########
# The light version (which i prefer, color fidelity remain high)

ConvertToYV16(interlaced=true) #
orig=last
ev=orig.assumetff().separatefields().selecteven()
od=orig.assumetff().separatefields().selectodd()
ev
ue_chroma = UToY(ev).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
ve_chroma = VToY(ev).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
YToUV(ue_chroma, ve_chroma)
MergeLuma(ev)
ev_filtered=last
od
uo_chroma = UToY(od).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
vo_chroma = VToY(od).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
YToUV(uo_chroma, vo_chroma)
MergeLuma(od)
od_filtered=last
interleave(ev_filtered,od_filtered)
assumefieldbased().assumetff().weave()

[kenator18]

Try this:

##Load filters:
LoadPlugin("C:\Program Files (x86)\AviSynth 2.6\plugins\KNLMeansCL-v1.1.1\KNLMeansCL.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth 2.6\plugins\TTempSmoothv094\TTempSmooth.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth 2.6\plugins\variableblur0.7_x86_x64\x86\variablebl ur.dll")
#####
MPEG2Source("C:\OIAM.d2v", cpu=0)
ConvertToYV16(interlaced=true) #
orig=last
ev=orig.assumetff().separatefields().selecteven()
od=orig.assumetff().separatefields().selectodd()
ev
ue_chroma = UToY(ev).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
ve_chroma = VToY(ev).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
YToUV(ue_chroma, ve_chroma)
MergeLuma(ev)
ev_filtered=last
od
uo_chroma = UToY(od).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
vo_chroma = VToY(od).blur(0,1.5).binomialblur(5).ttempsmooth(m axr=6,lthresh=150, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
YToUV(uo_chroma, vo_chroma)
MergeLuma(od)
od_filtered=last
interleave(ev_filtered,od_filtered)
assumefieldbased().assumetff().weave()

########
# The light version (which i prefer, color fidelity remain high)

ConvertToYV16(interlaced=true) #
orig=last
ev=orig.assumetff().separatefields().selecteven()
od=orig.assumetff().separatefields().selectodd()
ev
ue_chroma = UToY(ev).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
ve_chroma = VToY(ev).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
YToUV(ue_chroma, ve_chroma)
MergeLuma(ev)
ev_filtered=last
od
uo_chroma = UToY(od).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
vo_chroma = VToY(od).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=8)
YToUV(uo_chroma, vo_chroma)
MergeLuma(od)
od_filtered=last
interleave(ev_filtered,od_filtered)
assumefieldbased().assumetff().weave()

It seems like you found a way to remove most of it, merci !

[jagabo]

Almost as effective but much simpler:

MergeChroma(TTempSmooth(maxr=6, lthresh=150, cthresh=150, strength=8, interlaced=true))

[johnmeyer]

You can check out the rainbow removal I did for someone by clicking here:

https://forum.videohelp.com/threads/398864-Need-help-on-Avisynth#post2595644

I posted a visual before/after. I have used this same technique on other rainbow problems.

[Sharc]

Yes, SpotLess is a quite nifty tool (script). I don't know why it has been removed from the Avisynth wiki (or I just didn't find it there anymore).

[Selur]

It's listed under http://avisynth.nl/index.php/External_filters#Film_Damage_correction and links to https://forum.doom9.org/showthread.php?t=181777

[kenator18]

Thanks to all of you guys! I've tried using SpotLess which is quite slow and I find it very agressive. I probably need to take the time to tweak all the settings LOL

[Selur]

@ johnmeyer: Question regarding SpotLess and rainbow removal. Since you need to use a radT of 3+ how do you stop SpotLess to 'swallow' the wrong stuff?
when using

SpotLess(clip=clip, radT=4, ablksize=8, aoverlap=4, thsad=1000, thsad2=1000)

SpotLess often gets too destructive:



Cu Selur

[lollo]

Have a look here for delta mask using spotless https://forum.doom9.org/showthread.php?p=1944703#post1944703

[Selur]

Hmm,.. will do. Thanks.

[lollo]

You're welcome! johnmeyer was part of that wonderful discussion, so he can help in understanding if needed.

[johnmeyer]

The problem with any temporal filter is tuning them so they fix rather than ruin the video. The tuning parameters are almost always different, not only for different videos, but often from one scene to the next. This is why good restoration takes a long time.

The masking work was a noble effort, and I had high hopes it might work, but I never got a better result with the masks than without them.

[Alwyn]

Merge Chroma (via Jagabo, of course) did a great job on a rainbow issue I had.

[Selur]

The tuning parameters are almost always different, not only for different videos, but often from one scene to the next. This is why good restoration takes a long time.

Sadly I agree, for proper restoration you usually require (at least) a per scene filter approach which can take ages.

I never got a better result with the masks than without

sad to read that, but that saves me time trying to port it to Vapoursynth

Cu Selur

[lollo]

sad to read that, but that saves me time trying to port it to Vapoursynth

in that thread I read good results from user zorr and others...

[johnmeyer]

sad to read that, but that saves me time trying to port it to Vapoursynth

in that thread I read good results from user zorr and others...

Lollo makes a good point. Please don't take my comment as the last word on whether the masking approach worked, or not. Instead, do as he suggests and read through the thread to see what success other people had.

The reason I still have to do a lot of manual work when trying to use motion estimation to create intermediate frames (for either frame rate conversion or slow motion) is that the software which generates the mask on each frame has to sense when the motion estimation has broken down (like when panning across a line of closely-spaced fence posts). While the obnoxious visual artifacts are easy to see with your eye, it is really difficult to automatically construct a mask when those artifacts happen, and then have it cover the entire artifact.