Colorful horizontal bars/strips removal?

[HappyLee]

Hi, thanks a lot for helping me last time. Now I've met another strange source, that has colorful horizontal bars/strips everywhere (except those black areas). I've tried filters like DFMDeRainbow, Checkmate, DeScratch, but nothing works. Only MDegrain2 could make it looks much better, but still not enough.

Here are some snaps for the video source in 10 groups, the upper link is the original video source and the downer is what it looks like after strong MDegrain2. (The singer is Michael Jackson, and he's wearing a blue shirt and black pants.) Sorry that I can't access to picture sites like Photobucket for being Chinese, so I hope poco.cn would work out for you.

http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115639021.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115725067.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115751066.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115802027.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115815091.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115831061.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115926089.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/11/658182142013041611593909.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/11/6581821420130416115949096.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120016099.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120115088.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120129013.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120144044.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/12/658182142013041612015509.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120207037.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120218097.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120227054.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120239026.jpg

http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120250072.jpg
http://img142.poco.cn/mypoco/myphoto/20130416/12/6581821420130416120301067.jpg

I wonder if anyone has met a similar amateur film like that. I think if there's a script that can reduce 80% of those strips, MDegrain would do the rest. But the problem is, how to deal with them? The colorful strips are mainly red and green, so I'm thinking that RGBAdjust would reduce them a little bit after detecting them. I'm not an advanced programmer and have no solution, so I have to come and ask for help. And I'll be forever grateful to whoever would help me, thank you.

[Guest]

Where did you get the source video?

[HappyLee]

Where did you get the source video?

I got it from a 1988 Wembley performance amateur DVD. It's the only source known, and I don't know the holder of that VHS tape, but I think it's very important to enhance it.

[pandy]

Defreq? FFTQuiver? - perform operation on Cb and Cr plane separately.

[HappyLee]

Defreq? FFTQuiver? - perform operation on Cb and Cr plane separately.

I've just tried Defreq and it does no help at all. As for FFTQuiver, I couldn't load libFFTw3f-3.dll or FFTw3.dll for some reasons.

[StainlessS]

I've just tried Defreq and it does no help at all. As for FFTQuiver, I couldn't load libFFTw3f-3.dll or FFTw3.dll for some reasons.

Those are windows dll's not avisynth dll's. They go in system32 or SysWOW64 in 64 bit.

[HappyLee]

Those are windows dll's not avisynth dll's. They go in system32 or SysWOW64 in 64 bit.

Thanks. I've just tried FFTQuiver and it doesn't work, either. :(

[pandy]

Please provide sample of video + your script - where fftquiver and/or defreq not work.

[HappyLee]

Please provide sample of video + your script - where fftquiver and/or defreq not work.

OK. Here's a 13-second sample. It's not long but a good representive. Sorry that I tried about half an hour but couldn't upload to Mediafire, so I uploaded to Baidu (one of the biggest sites in China). It's easy to download even if you don't read Chinese.
http://pan.baidu.com/share/link?shareid=405398&uk=2684705515

My script is easy, mostly just a simple TomsMoComp for deinterlacing and Tweak(0,2.5,0,1.2) for improving the color. Thank you.

[pandy]

My script is easy, mostly just a simple TomsMoComp for deinterlacing and Tweak(0,2.5,0,1.2) for improving the color. Thank you.

FFTQuiver and/or Defreq - i need to see this part of your script.

Btw those stripes are related to limited big screen resolution recorded through camera with limited resolution - Moire pattern related to resampling of the sampled signal - perhaps you should consider this as a part of video itself - documentary and authentic - also it can be very difficult to remove such structure as camera make constant zoom and this is not static structure.

Quite good results have hqdn3d


Yadif(1,1).ConvertToYV12(Interlaced=false)
hqdn3d(ls=1, cs=150, lt=2, ct=1)
hqdn3d(ls=1, cs=100, lt=2, ct=2)
hqdn3d(ls=1, cs=50, lt=2, ct=3)


Defreq partially also can help - at a cost of time spent on searching for optimal settings there is a chance to improve this however everything can be not justified by source - chrominance bandwidth for typical VCR is around hundreds kHz - color resolution is around 20 - 50 lines max.


#DeFreq(plane=1, show=0, fx=0.05, fy=-8, dx=3, dy=7.25, sharp=75, fx2=0.05, fy2=8, dx2=3, dy2=7.25, sharp2=75)
#DeFreq(plane=2, show=0, fx=0.05, fy=-8, dx=3, dy=7.30, sharp=75, fx2=0.05, fy2=8, dx2=3, dy2=7.30, sharp2=75)
#DeFreq(plane=1, show=0, fx=0.05, fy=-22.75, dx=3, dy=7.25, sharp=75, fx2=0.05, fy2=23.5, dx2=3, dy2=7.25, sharp2=75)
#DeFreq(plane=2, show=0, fx=0.05, fy=-22.75, dx=3, dy=7.25, sharp=75, fx2=0.05, fy2=23.5, dx2=3, dy2=7.25, sharp2=75)

[HappyLee]

Quite good results have hqdn3d

Defreq partially also can help - at a cost of time spent on searching for optimal settings there is a chance to improve this however everything can be not justified by source - chrominance bandwidth for typical VCR is around hundreds kHz - color resolution is around 20 - 50 lines max.

Thank you so much! Defreq still doesn't make much difference to me, but hqdn3d does very well with those strips, although it could damage the color system when using too much at the same time. I find combining hqdn3d with MDegrain2 strongly effective:

hqdn3d(ls=0, cs=100, lt=0, ct=100)
super = MSuper(pel=2, sharp=1)
backward_vec2 = MAnalyse(super, isb = true, delta = 2, overlap=4)
backward_vec1 = MAnalyse(super, isb = true, delta = 1, overlap=4)
forward_vec1 = MAnalyse(super, isb = false, delta = 1, overlap=4)
forward_vec2 = MAnalyse(super, isb = false, delta = 2, overlap=4)
MDegrain2(super, backward_vec1,forward_vec1,backward_vec2,forward_vec2,thSAD=99999)


That works almost perfectly, except that you need to Tweak a little to fix the color. Thanks a lot for recommending hqdn3d to me. It's too bad that there haven't been a filter dealing with horizontal strips specially. :)

[pandy]

Thank you so much! Defreq still doesn't make much difference to me, but hqdn3d does very well with those strips, although it could damage the color system when using too much at the same time. I find combining hqdn3d with MDegrain2 strongly effective:

:::

That works almost perfectly, except that you need to Tweak a little to fix the color. Thanks a lot for recommending hqdn3d to me. It's too bad that there haven't been a filter dealing with horizontal strips specially. :)

Defreq or FFTQuiver are best methods to remove such problems but in this case nature of issue is very complex:

frequency is very random/variable (thus parameters for Defreq/FFTquiver shall be adjusted dynamically on adaptive fashion which is very difficult as those filters are optimized for static operation),

Source quality is problem (in VCR, chrominance is placed bellow luminance and provided bandwidth is highly limited when compared to luminance), in your case seem that noise from head switching is very visible, sometimes problem with servo tracking can be seen.

I still believe that with help of multiple entries (preferably adaptive) for Defreq/FFTQuiver they can show very good results but it will cost plenty of time where other method can be similarly effective and less demanding.