Color banding and noise removal

[cretindesalpes]

The artifacts don't come from DitherPost, the denoising stage is clearly the cause here... Try to keep a or h at its original value, or even both? I don't know exactly how these parameters affect the algorithm.

[cretindesalpes]

New Dither version (1.6), some important changes :

• Masktools 2 a48 is now the minimum requirement because some Dither functions use the new bitwise operators. I don't know how to automatically detect the Masktools version, so be sure to update your Masktools before installing this Dither version.
• Renamed the rem parameters into clsb in DitherPost and SmoothGrad functions
• Changed the default for the stacked (now true) and prot (now false) parameters.
• The modified MVTools2 have been synchronised with the latest official release (2.5.11.2 beta)
• Added various utility functions, mainly Dither_lut8, Dither_lut16, Dither_add16 and Dither_merge16.
• Some documentation improvements

[Heaud]

Great to see an update!

I have a question on the new utility functions. Would dither_convert_8_to_16 be able to replace the blank clip + stackvertical inside of gradfun3? Is there any other ways to improve gradfun3? I added a custom smoothgrad since I did not like the results of the orignal smoothgrad inside of the script (it was causing weird banding halos).

[cretindesalpes]

Originally Posted by Heaud

I have a question on the new utility functions. Would dither_convert_8_to_16 be able to replace the blank clip + stackvertical inside of gradfun3?

Yes, that's exactly what it does. But it's just a convenient way to write it in a script, it gives the same result and won't improve anything in GradFun3.

[Dogway]

I did a few dozen tests, tweaking sx,sy,bx,by,a and h, even removing MT(). None of them sort the artifact, furthermore I tried dither2pre to "gain" more accuracy but what I got were more artifacts, check the lineart. (h is quadrupled according.)

old-workflow
dither1pre
dither2pre

Im gonna test dither 1.6 now so I can tell you. edit: same results

[cretindesalpes]

Thank you very much for testing and helping me to improve the functions. I'm going to rewrite Dither1Pre to make the "slices" overlap, you'll tell me if it works better when it's ready.

[TheProfosist]

to me "Dither+SmoothGrad" seems like the best option. i would like to use it but cant determine from the help file what exactly you used/did. if it helps i usually denoise using fft3dgpu/fft3dfilter and sharpen using LSFmod before debanding.

[HymmexyLeadia]

im using a condensor mic to record samples, and theres a bit of noise in the background, i use audacity, but the noise removal algorithm isnt very good.
is there any software that does this better? thanks.
_________________
God rest ye, little children; let nothing you affright, For Jesus Christ, your Saviour, was born this happy night; Along the hills of Galilee the white flocks sleeping lay, When Christ, the Child of Nazareth, was born on Christmas day.
http://www.chung-sim.com/omron.3g3jv.html

[cretindesalpes]

Dither upgraded to v1.7. Changes:

• Slightly improved Dither1Pre and Dither2Pre functions, added the overlap and tvopt parameters.
• People complained about GradFun3 being too strong and washing fine shades away, so I changed the default parameter values for a better match with the gradfun2db default settings. thr is now set to 0.35 and radius to 12.

_____________________________________________________________________

Originally Posted by TheProfosist

to me "Dither+SmoothGrad" seems like the best option. i would like to use it but cant determine from the help file what exactly you used/did. if it helps i usually denoise using fft3dgpu/fft3dfilter and sharpen using LSFmod before debanding.

See: http://forum.doom9.org/showthread.ph...94#post1390594
but replace DitherPost (mode=2) with SmoothGrad (radius=16, thr=0.25, elast=2).DitherPost ()

[coolguy89]

Hi,
I tried your example above, but, the resolution changed from 720x480 to 720x240. Did I do something wrong?

[cretindesalpes]

You probably have a DitherPost() too many. Also make sure you have the latest Dither version. A few default parameters have changed recently and I updated the script in the post #8 accordingly.

[coolguy89]

This is what I use

FFT3DGpu (Sigma=1)
DitherPost (mode=2)

Function zzz_denoise (clip src, float "sigma", int "thr", bool "mask", int "sad")
{
	sigma = Default (sigma,    16)
	thr   = Default (thr,       5)
	mask  = Default (mask,  False)
	sad   = Default (sad,     200)

	w = src.Width ()
	h = src.Height ()

	# Motion analysis
	super   = MSuper (src)
	super_a = MSuper (src.TTempSmooth ().RemoveGrain (12))

	fwd_vect_3 = super_a.MAnalyse (isb=false, delta=3, overlap=4)
	fwd_vect_2 = super_a.MAnalyse (isb=false, delta=2, overlap=4)
	fwd_vect_1 = super_a.MAnalyse (isb=false, delta=1, overlap=4)
	bck_vect_1 = super_a.MAnalyse (isb=true,  delta=1, overlap=4)
	bck_vect_2 = super_a.MAnalyse (isb=true,  delta=2, overlap=4)
	bck_vect_3 = super_a.MAnalyse (isb=true,  delta=3, overlap=4)

	fwd_comp_2 = src.MCompensate (super, fwd_vect_2, thSAD=sad)
	fwd_comp_1 = src.MCompensate (super, fwd_vect_1, thSAD=sad)
	bck_comp_1 = src.MCompensate (super, bck_vect_1, thSAD=sad)
	bck_comp_2 = src.MCompensate (super, bck_vect_2, thSAD=sad)

	# Spatio-temporal denoising using modified dfttest
	c_dft = Interleave (fwd_comp_2, fwd_comp_1, src, bck_comp_1, bck_comp_2)
	c_dft = c_dft.dfttest (sigma=sigma, lsb=true)	# Double height
	c_dft = c_dft.SelectEvery (5, 2)

	# Temporal-only denoising using modified MDegrain
	c_deg = src.MDegrain3 (super, bck_vect_1, fwd_vect_1, bck_vect_2, fwd_vect_2, bck_vect_3, fwd_vect_3, thSAD=sad, lsb=true)	# Double height

	# Spatio-temporal denoising smoothes too much the details,
	# therefore we use pure temporal denoising on edges or detailed areas.
	edge_src = c_deg.Crop (0, 0, w, h)
	edge_mask = edge_src.mt_edge (mode="prewitt", thY1=thr, thY2=thr)
	edge_mask = edge_mask.mt_expand ()
	edge_mask = StackVertical (edge_mask, edge_mask)	# Double height
	c_hyb = mt_merge (c_dft, c_deg, edge_mask, luma=true, y=3, u=3, v=3)

	return (mask ? edge_mask.GreyScale () : c_hyb)
}

I'm using the latest Dither btw.

[cretindesalpes]

Use zzz_denoise(). This function does almost all the job, with SmoothGrad. DitherPost only converts the 16-bit result into a regular 8-bit dithered clip.

zzz_denoise()
SmoothGrad (radius=16, thr=0.25, elast=2)
DitherPost ()

If you want to keep FFT3DGpu as denoising filter, just do :

FFT3DGpu (Sigma=1)
GradFun3 ()

or

Dither2Pre (flt="FFT3DGpu(Sigma=4)")
SmoothGrad (radius=16, thr=0.25, elast=2)
DitherPost ()

[coolguy89]

It works! Thanks .

One more question, if I only want to deband, I only need the GradFun3, right? Or is there any other way in the dither packages that can improve debanding process.

[cretindesalpes]

Yes, use GradFun3 to deband only.

[coolguy89]

Thank you for your support and your awesome script.

[Dogway]

cretin, the error I was having was caused by MT inside dither1pre. I had to change splitvertical to true also to make it aware dither was changing the size.
On a note, before this I tried stacked=false and it gave me temporal artifacts. Finally after an afternoon battle it seems this code is what it works, no problems for encoding either.

MT("""Dither1Pre(flt="tnlmeans()",stacked=true)""",2,2,true)
DitherPost(stacked=true,prot=false)

You could also add it to documentation for reference.

[SSH4]

cretindesalpes can you check your mod16 of Mvtools?
Looks like all (latest mvtools-2.5.11.2-mod16a.zip and previous) version have some bug with tsadc. tsadc always == tsad.
you can check it if you made big tsad (about 600 or more) and try low tsadc. UV channel will be bad
on original mvtools-2.5.11.2 all work well.

thanx!

[cretindesalpes]

Indeed there was a bug, thanks for the report. I'll release a fixed version soon.

[cretindesalpes]

Dither 1.7.1 update. Changes:

• Dither.avsi doesn't use Max() anymore, so it can now run on Avisynth 2.5.7 or earlier.
• Bugfix in modified MVTools2: the thSADC argument was ignored in MDegrain1/2/3.