New approach to fine detail sharpening without edge halos. Very fast, HD also. - Page 4

feisty2 · 10th November 2014, 08:55

extract avsi files inside functions.7z and put them in your avisynth autoload folder
and run the script below

Code:

LWLibAVVideoSource("clip")
convert8to16 (tv_range=true)
EDIResize16(output="rgb48y",edimode="eedi3+nnedi3_repaired",noring=true,curve="linear")
#Dither_y_gamma_to_linear (tv_range_in=False, tv_range_out=False, curve="709",sigmoid=false)
#Dither_y_linear_to_gamma (tv_range_in=False, tv_range_out=False, curve="srgb",sigmoid=false)
YTo420_16()
soft=last
softo=soft

LWLibAVVideoSource("clip")
convert8to16 (tv_range=true)
converttoy8 ().tv2pc16 ()
#Dither_y_gamma_to_linear (tv_range_in=False, tv_range_out=False, curve="709",sigmoid=false)
#Dither_y_linear_to_gamma (tv_range_in=False, tv_range_out=False, curve="srgb",sigmoid=false)
YTo420_16()
search=last
searcho=search

soft
converttoy8 ()
LimitClip16(str=1.0, curve="srgb", tv_range=false)
LimitClip16(str=1.0, curve="srgb", tv_range=false)
YTo420_16()
limit=last

soft
converttoy8 ()
a=last
LimitClip16(str=1.0, curve="srgb", tv_range=false)
LimitClip16(str=1.0, curve="srgb", tv_range=false)
LimitClip16(str=1.0, curve="srgb", tv_range=false)
dither_sub16 (last,a,dif=true,y=3,u=3,v=3)
YTo420_16()
dif=last

soft
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superclip=last
superclipo=superclip

search
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
supersearch=last
supersearcho=supersearch

limit
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superlimit=last

dif
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superdif=last

soft=soft.selectevery (3,0)
dif=dif.selectevery (3,0)
limit=limit.selectevery (3,0)
pre=search
superclip=superclip.selectevery (3,0)
superdif=superdif.selectevery (3,0)
superlimit=superlimit.selectevery (3,0)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
r=last

soft=soft.selectevery (3,1)
dif=dif.selectevery (3,1)
limit=limit.selectevery (3,1)
pre=search
superclip=superclip.selectevery (3,1)
superdif=superdif.selectevery (3,1)
superlimit=superlimit.selectevery (3,1)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
g=last

soft=soft.selectevery (3,2)
dif=dif.selectevery (3,2)
limit=limit.selectevery (3,2)
pre=search
superclip=superclip.selectevery (3,2)
superdif=superdif.selectevery (3,2)
superlimit=superlimit.selectevery (3,2)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
b=last

soft=interleave (r,g,b)

search
converttoy8 ()
LimitClip16(str=1.0, curve="srgb", tv_range=false)
LimitClip16(str=1.0, curve="srgb", tv_range=false)
YTo420_16()
limit=last

search
converttoy8 ()
a=last
LimitClip16(str=1.0, curve="srgb", tv_range=false)
LimitClip16(str=1.0, curve="srgb", tv_range=false)
LimitClip16(str=1.0, curve="srgb", tv_range=false)
dither_sub16 (last,a,dif=true,y=3,u=3,v=3)
YTo420_16()
dif=last

limit
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superlimit=last

dif
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superdif=last

search=search.DeconvLimit16(pre=search, dif=dif, limit=limit, superclip=supersearch, superpre=supersearch, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)

soft
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
limit=last

soft
a=last
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
dither_sub16 (last,a,dif=true,y=3,u=3,v=3)
dif=last

soft
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superclip=last

search
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
supersearch=last

limit
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superlimit=last

dif
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superdif=last

soft=soft.selectevery (3,0)
dif=dif.selectevery (3,0)
limit=limit.selectevery (3,0)
pre=search
superclip=superclip.selectevery (3,0)
superdif=superdif.selectevery (3,0)
superlimit=superlimit.selectevery (3,0)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
r=last

soft=soft.selectevery (3,1)
dif=dif.selectevery (3,1)
limit=limit.selectevery (3,1)
pre=search
superclip=superclip.selectevery (3,1)
superdif=superdif.selectevery (3,1)
superlimit=superlimit.selectevery (3,1)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
g=last

soft=soft.selectevery (3,2)
dif=dif.selectevery (3,2)
limit=limit.selectevery (3,2)
pre=search
superclip=superclip.selectevery (3,2)
superdif=superdif.selectevery (3,2)
superlimit=superlimit.selectevery (3,2)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
b=last

limit=interleave (r,g,b)

search
ShrinkSharp16(original=searcho, str=1.0, tv_range=False)
ShrinkSharp16(original=searcho, str=1.0, tv_range=False)
limit=last

search
a=last
ShrinkSharp16(original=searcho, str=1.0, tv_range=False)
ShrinkSharp16(original=searcho, str=1.0, tv_range=False)
ShrinkSharp16(original=searcho, str=1.0, tv_range=False)
dither_sub16 (last,a,dif=true,y=3,u=3,v=3)
dif=last

limit
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superlimit=last

dif
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superdif=last

limit_s=search.DeconvLimit16(pre=search, dif=dif, limit=limit, superclip=supersearch, superpre=supersearch, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)

softo
Deconv16 (dif=true, tv_range=false)
dif=last

limit
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superlimit=last

dif
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superdif=last

soft=softo.selectevery (3,0)
dif=dif.selectevery (3,0)
limit=limit.selectevery (3,0)
pre=searcho
superclip=superclipo.selectevery (3,0)
superdif=superdif.selectevery (3,0)
superlimit=superlimito.selectevery (3,0)
superpre=supersearcho
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
r=last

soft=softo.selectevery (3,1)
dif=dif.selectevery (3,1)
limit=limit.selectevery (3,1)
pre=searcho
superclip=superclipo.selectevery (3,1)
superdif=superdif.selectevery (3,1)
superlimit=superlimito.selectevery (3,1)
superpre=supersearcho
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
g=last

soft=softo.selectevery (3,2)
dif=dif.selectevery (3,2)
limit=limit.selectevery (3,2)
pre=searcho
superclip=superclipo.selectevery (3,2)
superdif=superdif.selectevery (3,2)
superlimit=superlimito.selectevery (3,2)
superpre=supersearcho
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
b=last

soft=interleave (r,g,b)

searcho
Deconv16 (dif=true, tv_range=false)
dif=last

limit_s
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superlimit=last

dif
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superdif=last

search=searcho.DeconvLimit16(pre=searcho, dif=dif, limit=limit_s, superclip=supersearcho, superpre=supersearcho, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)

soft
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
limit=last

soft
a=last
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
ShrinkSharp16(original=softo, str=1.0, tv_range=False)
dither_sub16 (last,a,dif=true,y=3,u=3,v=3)
dif=last

soft
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superclip=last

search
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
supersearch=last

limit
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superlimit=last

dif
converttoy8 ()
GenSuperclip16(pel=4, tv_range=false, curve="srgb")
YTo420_16()
superdif=last

soft=soft.selectevery (3,0)
dif=dif.selectevery (3,0)
limit=limit.selectevery (3,0)
pre=search
superclip=superclip.selectevery (3,0)
superdif=superdif.selectevery (3,0)
superlimit=superlimit.selectevery (3,0)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
r=last

soft=soft.selectevery (3,1)
dif=dif.selectevery (3,1)
limit=limit.selectevery (3,1)
pre=search
superclip=superclip.selectevery (3,1)
superdif=superdif.selectevery (3,1)
superlimit=superlimit.selectevery (3,1)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
g=last

soft=soft.selectevery (3,2)
dif=dif.selectevery (3,2)
limit=limit.selectevery (3,2)
pre=search
superclip=superclip.selectevery (3,2)
superdif=superdif.selectevery (3,2)
superlimit=superlimit.selectevery (3,2)
superpre=supersearch
soft.DeconvLimit16(pre=pre, dif=dif, limit=limit, superclip=superclip, superpre=superpre, superdif=superdif, superlimit=superlimit, dct=1, str=1.00, tv_range=false)
b=last

interleave (r,g,b)

that's all of the "mcdeconvsharp" showed at #35

Hotte · 18th November 2014, 22:48

Summary of fine sharpening package for HD Material.

Thanks very much to feisty2 for his extensive help!

Step 1: Sharpen
The following simple but highly effective unsharp mask will emphasize only fine detail contrast for Full HD material and will add no edge halos if parameter w is set to low values. feisty2 suggested another very interesting approach with the dither-tool, but after extensive testing I came back to my first approach: It creates a tiny bit less halos / mid contrast, it is faster, it is simpler because it does not need dither. Code runs very fast. With multitasking put ahead you might process more than 35 fps (full HD) on a quadcore:

Code:

# Step 0: Multitasking options (Avisynth 2.6)

SetMemoryMax(1024)	    
SetMTMode(2,8)              # Try SetMTMode(3,8) if you run into troubles

# Step 1: Sharpen

AviSource("blabla.avi")
convertToYV12               # YV12 is needed by binomial blur  

s=100                           # strength of sharpen. If steps 2 and 3 are applied: try 80-100 (don´t go beyond). Step 1 only: try 40 to 70
w=4                             # width of sharpen. If steps 2 and 3 are applied: try 3-5. Step 1 only: try 1-2. Larger w = larger halos 

i=last.binomialBlur(varY=0.5, varC=0, Y=3, U=2, V=2,useMMX=true)
j=subtract(i,last).levels(0+s,1,255-s,0,255)
w=w*(128/s)
k=j.levels(127-w,1,128+w,127-w,128+w)
subtract(last,k)

Step 2: Remove edge halos
One optional single line helps to effectively remove remaining edge halos e.g. from preprocessing. Thanks Didier for this brilliant filter (that you should not forget to download from the avisynth pages). Setting supersampling (Parameter ss) to 1.5 rather than 1.0 will slow down processing-speed to 40% but helps a little with anti-aliasing (reduce stairs on linear structures).

Code:

DeHalo_alpha(highsens=100,ss=1.0)     # highsens adjusted to address larger halos and to nicely fit into the package

Step 3: Remove grain
This step slows down processing heavily (to some 4 fps) but helps a lot. Emphasizing fine detail might lead to fine grain flickering. Mdegrain averages out ("calms") fine grain using adjacent frames. There is no relevant drop of sharpness with that step. As additional gift Mdegrain also calms fine sensor chroma noise to make it considerably less obtrusive. Mdegrain also brings back a natural look to diffuse structures like clouds which could have been overstructured here and there by the sharpening process. Highly recommended!

Code:

super = MSuper(pel=2, sharp=1)
bv2 = MAnalyse(super, isb = true, delta = 2, overlap=4)
bv1 = MAnalyse(super, isb = true, delta = 1, overlap=4)
fv1 = MAnalyse(super, isb = false, delta = 1, overlap=4)
fv2 = MAnalyse(super, isb = false, delta = 2, overlap=4)
MDegrain2(super,bv1,fv1,bv2,fv2,thSAD=400)

You can do even more degraining by using MDegrain3 (see avisynth documentation) at the cost of even less processing speed, but I found that Mdegrain2 does a very good job already. Mdegrain3 can bring some very slight improvement to clips with strong sensor chroma noise.

Have fun with this package and thanks to all thread partcipants who had helped me so generously.

Hotte · 22nd November 2014, 10:05

Just to add one thing:

You can boooost detail sharpness by chaining more unsharp masks with lower sharpness settings together. Performance effects are almost irrelevant. It´s so fast.

s=100 w=4 like in Step 1 above gives me a well visible moderate detail sharpness amplification (in connection with dehalo_alpha and MDegrain2). If I go beyond 100 (up to the limit of 127) I am getting small whitish spots ("snow") instead of more sharpness if I zoom in to see what happens. Results will be a lot better and even sharper if I chain unsharp masks together. But be careful to choose low sharpness and width settings otherwise you either get something to shave your beard or the chain collapses to strange results. I found that going lower than s=30 or higher than s=70 will not behave like expected if masks are being chained together. Outside these boundaries the behaviour is not linear anymore. So you need to experiment with the parameters s and w to squeeze out the very best sharpness for your individual full-hd footage.

Exchanging Step 1 with the following lines gives me really excellent and well balanced detail sharpness, no halos, very few artefacts and calm grain (if MDregrain2 or 3 applied):

Code:

s=30
w=1
i=last.binomialBlur(varY=0.5, varC=0, Y=3, U=2, V=2, useMMX=true)
j=subtract(i,last).levels(0+s,1,255-s,0,255)
w=w*(128/s)
k=j.levels(127-w,1,128+w,127-w,128+w)
subtract(last,k)
s=30
w=1
i=last.binomialBlur(varY=0.5, varC=0, Y=3, U=2, V=2, useMMX=true)
j=subtract(i,last).levels(0+s,1,255-s,0,255)
w=w*(128/s)
k=j.levels(127-w,1,128+w,127-w,128+w)
subtract(last,k)

In special situations a third mask with s=30 and w=1 can add an extra kick to footage whose sharpness was underdeveloped right from the start.

See the detail views taken out of a 1920x1080 clip in the attachments. s=100 w=4 (Step 1), 2 times s=30 w=1 (chained), Edius NLE internal sharpener set to 28 and original compared.

SSH4 · 22nd November 2014, 11:35

http://avisynth.nl/index.php/Levels

Quote:

For adjusting brightness or contrast it is better to use Tweak or ColorYUV, because Levels also changes the chroma of the clip.

Also why just don't use faster Masktools that can make this code 3 times smaller, faster, more intuitive and that work without color conversion?

Hotte · 22nd November 2014, 11:46

I did not experience any relevant color shift with my clips.

Concerning performance Step 1 is very, very fast. 90% of processing time goes into (optional) MDegrain.

I´d be happy if you find things to improve, however I am not experienced enough to translate the code above into Masktools, Tweak or ColorYUV.

It would be great if you could help out with a code example - Thanks!

Hotte · 26th November 2014, 21:51

feisty2, should you listen:
could you tell me, if there is a parameter in "Function SharpLimit16 ()" to adjust the width (radius if you want) of sharpness, rather than the strength (represented by the str-parameter).
Thanks

GMJCZP · 27th November 2014, 00:33

I do not want to interrupt your conversation with feisty2, but I find it very interesting what you're doing and want to contribute something.

If you want to use a script that allows you to clean the image while sharpness I leave him this:
You could modify including DeHalo.

You can adjust the brightness and contrast try this:

YLevels (download masktools2 version)

SmoothAdjust

Fusion

feisty2 · 27th November 2014, 01:26

the function doesn't have such parameter, you can set radius on your sharpen filter
its not a sharpen script, it only limits the sharpen result

Hotte · 27th November 2014, 08:23

GMJCZP, thanks for your feedback. Especially YLevels looks interesting as a replacement for levels. Could improve high sharpening levels - maybe. Also I´ll have a look what Didier´s script is good for.

I will give it a try at the next opportunity. Hopefully YLevels behaves like levels, otherwise it will be getting difficult since I only have a basic understanding what the code does - but not in deep.

The other commands I honestly do not know where they could add something to the issue. I do not really adjust contrast nor brightness. It's just an unsharp mask.

feisty2 - got the idea now. Well it actually does sharpen - and not too bad, really! Even though it was not meant for that - funny.

I am still really fascinated what these few codelines do to my videos. Looking at MDegrain alone: The results come next to a miracle.

Hotte · 28th November 2014, 00:02

GMJCZP, I tried ylevels: I replaced the two levels-statements with ylevels leaving all the parameters as they were.

Sharpen works with ylevels, but differences are big: With ylevels sharpening affects more mid-detail rather than fine-detail and creates much stronger halos with my full-HD footage. On the chroma side there was no visible improvement.

I don´t know what it is in this old code of Step 1 but with no other solution I tried before have I seen this special micro-contrast enhancement affecting only fine structures and giving a perception I would describe as 'discreet bottom-up' sharpness - for anyone who likes it - of course.

More comments and suggestions are welcome any time.

SSH4 · 28th November 2014, 11:58

Quote:

Originally Posted by Hotte

I did not experience any relevant color shift with my clips.

Concerning performance Step 1 is very, very fast. 90% of processing time goes into (optional) MDegrain.

I´d be happy if you find things to improve, however I am not experienced enough to translate the code above into Masktools, Tweak or ColorYUV.

It would be great if you could help out with a code example - Thanks!

Looks like you try implement something like this:
Limit amount of overshoot and undershoot from Sharp filter.

Code:

src=last
overshoot=4
undershoot=16
blr=0.5

i=src.binomialBlur(varY=blr, varC=0, Y=3, U=2, V=2, useMMX=true)
a=mt_makediff(i,src,u=2,v=2,chroma="128")                                      #equal your substract()
x=mt_makediff(src,a, y=3, U=3, V=3)                                               #equal your second substract() that create Sharp
mt_clamp(x,src,src,overshoot,undershoot)                                         #this line limit amount of previous step no more than +4 and not less than -16 from original pixel value

GMJCZP · 28th November 2014, 14:57

Quote:

Originally Posted by Hotte

GMJCZP, I tried ylevels: I replaced the two levels-statements with ylevels leaving all the parameters as they were.

Sharpen works with ylevels, but differences are big: With ylevels sharpening affects more mid-detail rather than fine-detail and creates much stronger halos with my full-HD footage. On the chroma side there was no visible improvement.

I don´t know what it is in this old code of Step 1 but with no other solution I tried before have I seen this special micro-contrast enhancement affecting only fine structures and giving a perception I would describe as 'discreet bottom-up' sharpness - for anyone who likes it - of course.

You should note that Ylevels truly is a set of 4 functions. Must try which one fits your needs.

Hotte · 30th November 2014, 21:28

GMJCZP, I tried all the ylevel-functions now. Sorry to tell that all show similar characteristics concerning unsharp mask. They will all show much more halos and more mid-detail rather than fine detail enhancement, so stay behind in terms of fine-sharpening. But anyway thanks for your feedback.

Hotte · 30th November 2014, 21:35

SSH4, are your codelines a sharpener or just something to limit sharpness ? I tried with several overshoot/undershoot parameters but did not see any significant sharpening effect. Thanks.

GMJCZP · 30th November 2014, 23:45

Then try SmoothAdjust or Fusion.

18th November 2014, 22:48	#62 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	Summary of fine sharpening package for HD Material. Thanks very much to feisty2 for his extensive help! Step 1: Sharpen The following simple but highly effective unsharp mask will emphasize only fine detail contrast for Full HD material and will add no edge halos if parameter w is set to low values. feisty2 suggested another very interesting approach with the dither-tool, but after extensive testing I came back to my first approach: It creates a tiny bit less halos / mid contrast, it is faster, it is simpler because it does not need dither. Code runs very fast. With multitasking put ahead you might process more than 35 fps (full HD) on a quadcore: Code: # Step 0: Multitasking options (Avisynth 2.6) SetMemoryMax(1024) SetMTMode(2,8) # Try SetMTMode(3,8) if you run into troubles # Step 1: Sharpen AviSource("blabla.avi") convertToYV12 # YV12 is needed by binomial blur s=100 # strength of sharpen. If steps 2 and 3 are applied: try 80-100 (don´t go beyond). Step 1 only: try 40 to 70 w=4 # width of sharpen. If steps 2 and 3 are applied: try 3-5. Step 1 only: try 1-2. Larger w = larger halos i=last.binomialBlur(varY=0.5, varC=0, Y=3, U=2, V=2,useMMX=true) j=subtract(i,last).levels(0+s,1,255-s,0,255) w=w(128/s) k=j.levels(127-w,1,128+w,127-w,128+w) subtract(last,k) Step 2:* Remove edge halos One optional single line helps to effectively remove remaining edge halos e.g. from preprocessing. Thanks Didier for this brilliant filter (that you should not forget to download from the avisynth pages). Setting supersampling (Parameter ss) to 1.5 rather than 1.0 will slow down processing-speed to 40% but helps a little with anti-aliasing (reduce stairs on linear structures). Code: DeHalo_alpha(highsens=100,ss=1.0) # highsens adjusted to address larger halos and to nicely fit into the package Step 3: Remove grain This step slows down processing heavily (to some 4 fps) but helps a lot. Emphasizing fine detail might lead to fine grain flickering. Mdegrain averages out ("calms") fine grain using adjacent frames. There is no relevant drop of sharpness with that step. As additional gift Mdegrain also calms fine sensor chroma noise to make it considerably less obtrusive. Mdegrain also brings back a natural look to diffuse structures like clouds which could have been overstructured here and there by the sharpening process. Highly recommended! Code: super = MSuper(pel=2, sharp=1) bv2 = MAnalyse(super, isb = true, delta = 2, overlap=4) bv1 = MAnalyse(super, isb = true, delta = 1, overlap=4) fv1 = MAnalyse(super, isb = false, delta = 1, overlap=4) fv2 = MAnalyse(super, isb = false, delta = 2, overlap=4) MDegrain2(super,bv1,fv1,bv2,fv2,thSAD=400) You can do even more degraining by using MDegrain3 (see avisynth documentation) at the cost of even less processing speed, but I found that Mdegrain2 does a very good job already. Mdegrain3 can bring some very slight improvement to clips with strong sensor chroma noise. Have fun with this package and thanks to all thread partcipants who had helped me so generously. Last edited by Hotte; 18th November 2014 at 22:50.

22nd November 2014, 10:05	#63 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	Just to add one thing: You can boooost detail sharpness by chaining more unsharp masks with lower sharpness settings together. Performance effects are almost irrelevant. It´s so fast. s=100 w=4 like in Step 1 above gives me a well visible moderate detail sharpness amplification (in connection with dehalo_alpha and MDegrain2). If I go beyond 100 (up to the limit of 127) I am getting small whitish spots ("snow") instead of more sharpness if I zoom in to see what happens. Results will be a lot better and even sharper if I chain unsharp masks together. But be careful to choose low sharpness and width settings otherwise you either get something to shave your beard or the chain collapses to strange results. I found that going lower than s=30 or higher than s=70 will not behave like expected if masks are being chained together. Outside these boundaries the behaviour is not linear anymore. So you need to experiment with the parameters s and w to squeeze out the very best sharpness for your individual full-hd footage. Exchanging Step 1 with the following lines gives me really excellent and well balanced detail sharpness, no halos, very few artefacts and calm grain (if MDregrain2 or 3 applied): Code: s=30 w=1 i=last.binomialBlur(varY=0.5, varC=0, Y=3, U=2, V=2, useMMX=true) j=subtract(i,last).levels(0+s,1,255-s,0,255) w=w(128/s) k=j.levels(127-w,1,128+w,127-w,128+w) subtract(last,k) s=30 w=1 i=last.binomialBlur(varY=0.5, varC=0, Y=3, U=2, V=2, useMMX=true) j=subtract(i,last).levels(0+s,1,255-s,0,255) w=w(128/s) k=j.levels(127-w,1,128+w,127-w,128+w) subtract(last,k) In special situations a third mask with s=30 and w=1 can add an extra kick to footage whose sharpness was underdeveloped right from the start. See the detail views taken out of a 1920x1080 clip in the attachments. s=100 w=4 (Step 1), 2 times s=30 w=1 (chained), Edius NLE internal sharpener set to 28 and original compared. Attached Images Last edited by Hotte; 22nd November 2014 at 10:32.

27th November 2014, 00:33	#67 \| Link
GMJCZP Registered User Join Date: Apr 2010 Location: I have a statue in Hakodate, Japan Posts: 739	I do not want to interrupt your conversation with feisty2, but I find it very interesting what you're doing and want to contribute something. If you want to use a script that allows you to clean the image while sharpness I leave him this: You could modify including DeHalo. You can adjust the brightness and contrast try this: YLevels (download masktools2 version) SmoothAdjust Fusion Last edited by GMJCZP; 27th November 2014 at 00:38.

27th November 2014, 08:23	#69 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	GMJCZP, thanks for your feedback. Especially YLevels looks interesting as a replacement for levels. Could improve high sharpening levels - maybe. Also I´ll have a look what Didier´s script is good for. I will give it a try at the next opportunity. Hopefully YLevels behaves like levels, otherwise it will be getting difficult since I only have a basic understanding what the code does - but not in deep. The other commands I honestly do not know where they could add something to the issue. I do not really adjust contrast nor brightness. It's just an unsharp mask. feisty2 - got the idea now. Well it actually does sharpen - and not too bad, really! Even though it was not meant for that - funny. I am still really fascinated what these few codelines do to my videos. Looking at MDegrain alone: The results come next to a miracle. Last edited by Hotte; 28th November 2014 at 00:06.

28th November 2014, 00:02	#70 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	GMJCZP, I tried ylevels: I replaced the two levels-statements with ylevels leaving all the parameters as they were. Sharpen works with ylevels, but differences are big: With ylevels sharpening affects more mid-detail rather than fine-detail and creates much stronger halos with my full-HD footage. On the chroma side there was no visible improvement. I don´t know what it is in this old code of Step 1 but with no other solution I tried before have I seen this special micro-contrast enhancement affecting only fine structures and giving a perception I would describe as 'discreet bottom-up' sharpness - for anyone who likes it - of course. More comments and suggestions are welcome any time. Last edited by Hotte; 28th November 2014 at 00:08.

22nd November 2014, 11:46	#65 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	I did not experience any relevant color shift with my clips. Concerning performance Step 1 is very, very fast. 90% of processing time goes into (optional) MDegrain. I´d be happy if you find things to improve, however I am not experienced enough to translate the code above into Masktools, Tweak or ColorYUV. It would be great if you could help out with a code example - Thanks!

26th November 2014, 21:51	#66 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	feisty2, should you listen: could you tell me, if there is a parameter in "Function SharpLimit16 ()" to adjust the width (radius if you want) of sharpness, rather than the strength (represented by the str-parameter). Thanks

27th November 2014, 01:26	#68 \| Link
feisty2 I'm Siri Join Date: Oct 2012 Location: void Posts: 2,633	the function doesn't have such parameter, you can set radius on your sharpen filter its not a sharpen script, it only limits the sharpen result

30th November 2014, 21:28	#73 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	GMJCZP, I tried all the ylevel-functions now. Sorry to tell that all show similar characteristics concerning unsharp mask. They will all show much more halos and more mid-detail rather than fine detail enhancement, so stay behind in terms of fine-sharpening. But anyway thanks for your feedback.

30th November 2014, 21:35	#74 \| Link
Hotte Registered User Join Date: Oct 2014 Posts: 209	SSH4, are your codelines a sharpener or just something to limit sharpness ? I tried with several overshoot/undershoot parameters but did not see any significant sharpening effect. Thanks.

30th November 2014, 23:45	#75 \| Link
GMJCZP Registered User Join Date: Apr 2010 Location: I have a statue in Hakodate, Japan Posts: 739	Then try SmoothAdjust or Fusion.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode