DegrainDenoiseSharpen

[ingoldie]

Can you help me on improving the following script to get an effective result about denoising without loss of detail?

It causes loss of detail when I increase the value ​​of the denoising parameters(sigma,thSAD etc.).
But I can't get an effective result about denoising when I don't increase the value of the denoising parameters.

function DegrainDenoiseSharpen(clip c)
{
super = MSuper(c, pel=2, sharp=1)
mvbw2 = super.MAnalyse(isb = true, delta = 1, blksize=4, overlap=2)
mvbw1 = super.MAnalyse(isb = false, delta = 1, blksize=4, overlap=2)
mvfw1 = super.MAnalyse(isb = false, delta = 1, blksize=4, overlap=2)
mvfw2 = super.MAnalyse(isb = false, delta = 1, blksize=4, overlap=2)
den= MDegrain2(c.sharpen(0.57).FFT3DFilter(sigma=1.6, dehalo=0.4), super, mvbw1, mvfw1, mvbw2, mvfw2, thSAD=150)
return den
}

[guest]

Can you help me on improving the following script to get an effective result about denoising without loss of detail?

It causes loss of detail when I increase the value ​​of the denoising parameters(sigma,thSAD etc.).
But I can't get an effective result about denoising when I don't increase the value of the denoising parameters.

function DegrainDenoiseSharpen(clip c)
{
super = MSuper(c, pel=2, sharp=1)
mvbw2 = super.MAnalyse(isb = true, delta = 1, blksize=4, overlap=2)
mvbw1 = super.MAnalyse(isb = false, delta = 1, blksize=4, overlap=2)
mvfw1 = super.MAnalyse(isb = false, delta = 1, blksize=4, overlap=2)
mvfw2 = super.MAnalyse(isb = false, delta = 1, blksize=4, overlap=2)
den= MDegrain2(c.sharpen(0.57).FFT3DFilter(sigma=1.6, dehalo=0.4), super, mvbw1, mvfw1, mvbw2, mvfw2, thSAD=150)
return den
}

I would suggest you have a look at Dogways thread as he has many many filters that work really good.

[Dogway]

Use a nice prefilter and then moderate denoise settings in SMDegrain. I don't know the source so can't help further.

[DTL]

MDegrain2 uses really small tr of 2. If you need more degrain better to use MDegrainN with much larger tr of 5..10..15 or more. If you process sort of photon shot noise with Poisson/Gauss distribution its amplitude decreases about as SQRT((2_x_tr) + 1).
So for tr=2 you got max possible degrain of about sqrt(5)=2.2 times less of input.

With tr=16 it is about sqrt(33)=5.7 times smaller (if all blocks fit in averaging process well - SAD good below thSAD).

You can look into EZdenoise https://forum.doom9.org/showthread.php?t=183192 for syntax with MDegrainN with variable tr-value.