Old clip - need some help

[salehin]

I'm thinking of encoding this clip (http://www.sendspace.com/file/5h2hn1) with either of the following avs scripts:
SetMemoryMax(128)
DGDecode_mpeg2source("M:\CALCUTTA\VIDEO_TS\1969.d2v",info=3)
ColorMatrix(hints=true)
crop(30, 14, -30, -18)

dull=last.Spline36Resize(656,448)
sharp = dull.LimitedSharpenfaster(dest_x=656,dest_y=448,Strength=100)
Soothe(sharp,dull,45)
clip1=DGDecode_mpeg2source("M:\CALCUTTA\VIDEO_TS\1969.d2v",info=3).ColorMatrix(hints=true)

a=clip1.crop(30, 14, -30, -18)
clip2=DeGrainMedian(a, limitY=2,limitUV=3,mode=1).Spline36Resize(656, 448)

GrainOptimizer(clip2).LimitedSharpenFaster(Strength=150)

I was wondering if I could improve it further (other filters and values)- would be a great help if you could kindly post the script - i'm not good at all). Obviously I don't want over-sharpened encode.

Thanks :)

[Adub]

You are definitely going to want to deal with the grain before you any sharpening.

Degainmedian is a good start, but look into Removgrain/dirt and fft3dfilter()/fft3dgpu(). They will be able to remove a more "full" amount of noise, and deal with the dirt and crap that is all over the clip.

Also, try this thread:
http://forum.doom9.org/showthread.php?p=1093821#post1093821

If I have time this weekend, I will give it a run, but not anytime sooner.

[salehin]

Thanks for the tips merlin. I tried withRemovgrain/dirt and fft3dfilter()/fft3dgpu()

o RemoveDirt: Don't know how to use it's syntax

o fft3dfilter: gives an error Evaluation: Syste, exception- access violation"

o fft3dgpu: works, but after few previews, AvsP/the filter tells me that I need more video memory for this specfic filter (mine is nvidia 64 MB for Geforce 5200, laptop). But the preview looked better than before with the following settings:

SetMemoryMax(128)
clip1=DGDecode_mpeg2source("M:\CALCUTTA\VIDEO_TS\VTS_02_1.d2v",info=3).ColorMatrix(hints=true)

a=clip1.crop(30, 14, -30, -18)
clip2=fft3dfilter(a, degrid=1, sigma=2.0)
#clip2=fft3dgpu(a, sigma=2.0)
GrainOptimizer(clip2).LimitedSharpenFaster(Strength=50).Spline36Resize(656, 448)

[Pookie]

It doesn't need that much processing - deflicker and reduction of the dancing grain will probably be enough. There isn't much detail to sharpen. Just stabilize it and it will be more pleasing to the eyes.

Try the deflicker plugin from Fizick and have a look at this script -


http://forum.doom9.org/showthread.php?p=1073349#post1073349

[Adub]

Yeah, I will give you some more feedback tomorrow, but I have midterms to study for right now. So far, I have found that fft3dfilter does wonders on this clip.

Combining it with Seesaw at a low setting (or LimitedSharpenFaster) can really start to make it look good. I will post some settings later.

[Adub]

Okay, I spent a little time with your clip, and found that FFT3dGPU works wonders on most of it.

Here are the filters that I used. You should adjust them to your satisfaction as they were just ruff estimates. Especially the sharpening. Tweak that until you find a balance that you like between blur and halo's.

Crop(30,14,-30,-18)
Lanczos4Resize(656,448)

c = last
mdata = DePanEstimate(c)
DePanInterleave(c, data=mdata)
SelectEvery(3, 1)
Deflicker(percent=40, lag=30,scene=15, info=false)

FFT3dGPU(sigma=1.1, beta=1.0, plane=0,sharpen=.7)

Tweak(0, 1.1, 3)


Feel free to move the crop and Lanczos resize to after all filtering has been done if you want. This is actually what I had done, but it makes more sense to crop and resize before filtering is applied, as it runs slightly faster.

Now, the two most important values that you will want to change (if at all) will be the sigma and sharpen values of FFT3dGPU.

Have fun!
If you need more help, just post and I will do the best I can, in addition to anyone else who has input.

[salehin]

thanks a lot, merlin. will try some values around those :)

[Sagekilla]

Personally I'd recommend you use Spline36Resize or Spline16Resize instead of Lanczos4Resize. They tend to be much easier to compress (you'd be surprised) and in crf mode of x264, use less bitrate. Even with small resizes (720x480 to 864x480) there is a noticeable difference.