Blocks on black?

[noelle675]

Ok, I have no idea what this problem is called so blindly trying filters is getting me no where. Could any give me a name or a solution to get rid of the blocks on the hair? A lot of my blacks usually have this on them...


Original

I've tried deblock, dfftest and temporal cleaner, but that doesn't seem to work quite right, or maybe I just don't know what luma/chroma variable I need. I'm still trying to figure out all the variables for everything.

I've gotten it so far to look like this but it's still not gone yet.

Code: Select all

ConvertToYV12()
deen("a2d",4,9,11)
msharpen(threshold=15,strength=150)
deblock()

Thank you in advance for any advice. Yukimura's hair thanks you, too.

[Scintilla]

The hair (that is the problem area, right?) in the second picture looks decent, and I wouldn't reasonably expect to be able to do much better. The only problem is that you're smearing details in the rest of the picture.

I don't really know if this is the kind of noise that a deblocker will help you with; I'm fairly sure that deblockers are supposed to help more with this kind of blocking.
You may want to try VagueDenoiser; I think the noise in my sample image is similar to your kind of noise, just in a lighter area. Also, VagueDenoiser shouldn't smear details as much as Deen.

[Qyot27]

I had to crank my monitor's brightness up to the maximum just to see the problem in that first picture. Without really knowing much beyond what could be causing it, I'm guessing gradiation of a sort, in which case I'd hit it with gradfun2db(), with a value of 2.0 - or maybe more, if it really seemed to need it.

One comment, though, is that those deen settings are way too high. There's very few sources I've encountered that are truly bad enough to warrant a 4,9,11 assault, no matter what mode you're using (I stick to w3d mode).

[Qyot27]

Without really knowing much about what could be causing it

Fixed. Also, the link to gradfun2db:
http://forum.doom9.org/showthread.php?t=108681

[Kariudo]

Just looks like quantization noise from compressing, a temporal smoother might help out some (that type of noise fluctuates with a relatively large change in color value)...fluxsmoothST() is worth a shot.

[mirkosp]

Without really knowing much about what could be causing it

Fixed. Also, the link to gradfun2db:
http://forum.doom9.org/showthread.php?t=108681

gradfun2db is good times, but be sure to grab the gradfun2dbmod OR paste the gradfunkmirror() code into an avsi. Its code is the following:

Code: Select all

Function gradfunkmirror( clip c, float "strength" )
{
	strength	=	default( strength, 1.2 )
	w			=	c.width()
	h			=	c.height()
	vflip		=	c.FlipVertical()
	hflip		=	c.FlipHorizontal()
	
	stackhorizontal( hflip.crop( w-16, 0, 16, h ).addborders( 0, 16, 0, 16 ),
	\	stackvertical( vflip.crop( 0, h-16, w, 16 ), c, vflip.crop( 0, 0, w, 16 ) ),
	\	hflip.crop( 0, 0, 16, h ).addborders( 0, 16, 0, 16 ) )
	gradfun2db( strength )
	crop( 16, 16, -16, -16 )

	Return last
}

I'm telling you to do so because the original gradfun2db has an issue that causes the 16 margin borders to NOT get debanded.

[noelle675]

It worked! Thank you all so much. It looks pretty close to perfect. Can't really ask for much more than that.

Code: Select all

ConvertToYV12()
Deen(mode="a2d", rad=4, thrY=7, thrUV=9, min=0.5)
VagueDenoiser(threshold=2, method=3, nsteps=6, chromaT=2.0)
gradfun2db(3)
msharpen(threshold=15,strength=120)
Function GradFunkMirror( clip c, float "strength" )
{
#counters edge effects for gradfun2db()
#gradfun2db() requires width mod8 and height mod2
strength = default( strength, 2)
w = c.width()
h = c.height()
vflip = c.FlipVertical()
hflip = c.FlipHorizontal()

stackhorizontal( hflip.crop( w-16, 0, 16, h ).addborders( 0, 16, 0, 16 ),
\ stackvertical( vflip.crop( 0, h-16, w, 16 ), c, vflip.crop(0, 0, w, 16 ) ),
\ hflip.crop( 0, 0, 16, h ).addborders( 0, 16, 0, 16 ) )
gradfun2db( strength )
crop( 16, 16, -16, -16 )

Return last
}

The hair (that is the problem area, right?) in the second picture looks decent, and I wouldn't reasonably expect to be able to do much better. The only problem is that you're smearing details in the rest of the picture.

I don't really know if this is the kind of noise that a deblocker will help you with; I'm fairly sure that deblockers are supposed to help more with this kind of blocking.
You may want to try VagueDenoiser; I think the noise in my sample image is similar to your kind of noise, just in a lighter area. Also, VagueDenoiser shouldn't smear details as much as Deen.

Oh, the background is actually permanently like that. I should have mentioned that earlier. That's perfect. I've changed the settings on Deen and added VagueDenoiser and it's working ALOT better with my other parts with a detailed backgrounds. Thank you very much!

I had to crank my monitor's brightness up to the maximum just to see the problem in that first picture. Without really knowing much beyond what could be causing it, I'm guessing gradiation of a sort, in which case I'd hit it with gradfun2db(), with a value of 2.0 - or maybe more, if it really seemed to need it.

One comment, though, is that those deen settings are way too high. There's very few sources I've encountered that are truly bad enough to warrant a 4,9,11 assault, no matter what mode you're using (I stick to w3d mode).

Ah, yes!! That's it. Gradients. I never thought that would be what was causing it. That fixed it perfectly. And as an added bonus, you also helped me fix the light coming off of the moon. I'll definitely have to remember that one.

Spoiler :

oh are the settings too high? I have to admit I'm still trying to wrap my brain around the different values and how to actually use them right. I took another look at Scintilla's guide and she has the values labeled. That makes things a lot easier to for me. Thanks for pointing out that they were too high.

Just looks like quantization noise from compressing, a temporal smoother might help out some (that type of noise fluctuates with a relatively large change in color value)...fluxsmoothST() is worth a shot.

Ah, so that's how that one works. It didn't work too well with this problem, but I'm sure it'll come in handy later on now that I actually know what it does. Thanks.

I'm telling you to do so because the original gradfun2db has an issue that causes the 16 margin borders to NOT get debanded.

Perfect. Thank you!

[Qyot27]

Well, it's good there was some improvement, but when I said that deen's settings were too high, I meant that using 4 for your radius, especially with high values for the other two, was overkill. I rarely ever go above deen("w3d",3,8,10) as an upper limit, and usually use it somewhere between 2,8,8 and 3,6,8 - 2,8,8 being more common by far. If I want temporal smoothing in there too, I extend it out to deen("w3d",2,8,8,8,8) - the thing to understand about filtering is that how much to do depends on how good or bad the source footage is.

Most shows produced the last 8 or 9 years are nowhere near bad enough to require heavy smoothing or even heavy sharpening (some of the U.S. domestic releases of the 2001-2004 period might, though). And that kind of stuff will harm the source even more than it might seem to help. It's like attacking the footage with a meat cleaver when what you really need is a scalpel.

If you're wanting to use gradfunkmirror, you don't call it like that. Copy and paste the function into a separate file, gradfunkmirror.avsi, and stick that into your plugins folder, then use it like you would any other filter. The way you have the script right now means you're not even using gradfunkmirror, because you aren't calling it.

[Scintilla]

I took another look at Scintilla's guide and she has the values labeled.

I swear, this never gets old!

[noelle675]

Well, it's good there was some improvement, but when I said that deen's settings were too high, I meant that using 4 for your radius, especially with high values for the other two, was overkill. I rarely ever go above deen("w3d",3,8,10) as an upper limit, and usually use it somewhere between 2,8,8 and 3,6,8 - 2,8,8 being more common by far. If I want temporal smoothing in there too, I extend it out to deen("w3d",2,8,8,8,8) - the thing to understand about filtering is that how much to do depends on how good or bad the source footage is.

Most shows produced the last 8 or 9 years are nowhere near bad enough to require heavy smoothing or even heavy sharpening (some of the U.S. domestic releases of the 2001-2004 period might, though). And that kind of stuff will harm the source even more than it might seem to help. It's like attacking the footage with a meat cleaver when what you really need is a scalpel.

If you're wanting to use gradfunkmirror, you don't call it like that. Copy and paste the function into a separate file, gradfunkmirror.avsi, and stick that into your plugins folder, then use it like you would any other filter. The way you have the script right now means you're not even using gradfunkmirror, because you aren't calling it.

Oh. I had no idea that's how you use it. There we go. Here's what I have now.

Code: Select all

ConvertToYV12()
Deen(mode="a2d", rad=4, thrY=5, thrUV=9, min=0.5)
VagueDenoiser(threshold=2, method=3, nsteps=6, chromaT=2.0)
msharpen(threshold=15,strength=120)
GradFunkMirror(3)

Well, it's good there was some improvement, but when I said that deen's settings were too high, I meant that using 4 for your radius, especially with high values for the other two, was overkill. I rarely ever go above deen("w3d",3,8,10) as an upper limit, and usually use it somewhere between 2,8,8 and 3,6,8 - 2,8,8 being more common by far. If I want temporal smoothing in there too, I extend it out to deen("w3d",2,8,8,8,8) - the thing to understand about filtering is that how much to do depends on how good or bad the source footage is.

I feel really bad asking, but if you wouldn't mind could you explain what "radius" and "threshold" are? I tried changing the radius and the two threshold values with the code I have now just to see what happens and I'm not really seeing much of a difference. I read bits of Scintilla guides again to try and get my head to osmosis the knowledge in, but the farthest my brain got was that thrY and thrUV deal with chroma and luma. So now I don't know what either of those are. The best guess is chroma has to do with color and luma has to do with light, but that's vague and not helpful. I did notice some minor noise like smoothing on another part of the clip that is got rid of.

I took another look at Scintilla's guide and she has the values labeled.

I swear, this never gets old!

You mean I'm not the first idiot you've enlightened by naming values? And while you're helping I got to thank you for your wonderful guides. They're extremely helpful.