Best Mosquito removing trick!

[birdy]

I wonder what is your best mosquito removing trick?!?

My movies almost look great, anyway I am satisfied with the quality I get. But whats really enoying is this mosquito noise that has been discused so often. Well it seems that we have to get used to it with if using xvid, but I am also sure that many have there ways of minimizing or removing them as much as possible.

So I was wondering if you would like to share the info!

What's best filter, setting etc to do that in your opinion?

[Morbo]

Lowering the resolution helps if its a complicated source.

Absolute lowest Ive used is,512x288....

Any lower and you get "steps" from resize.

My 2cents anyways:sly:

Cheers!!

[wotef]

i find this is the BEST for mosquitos, you can trust me cause i am l33t sbc nandub anime fansubbaaaaa and i can compress a 4 hour movie into 175mb @ full D1 resolution, oh yes, all without filters, too cause i like a sharp image and no i'm sorry i can't tell you my filterchain cause that's only for me and my ninja homiez, pffffft....

http://www.hawpar.com/images/review_healthcare_pic1.jpg

p.s. don't diss me and my mad skillz or i'll get my dad to beat you up

[iago]

@wotef

LOL :D

Btw, what about this:

"UnDot is a simple median filter for removing dots, that is stray orphan pixels and mosquito noise. It basicly just clips each pixel value to stay within min and max of its eight surrounding neigbors."
(from the readme of Tom's UnDot filter)

Or:

TemporalSoften(1,5,0) which is MoonWalker's long time favourite ;).

[birdy]

haha wotef :)

@iago,
thanx again !

[MrBunny]

@birdy

Are you talking about mosquito noise caused by the xvid compression or mosquito noise in the source?
If the former, with all respects to iago :) , I don't think undot would help with that (as it takes place before the encode step). Probably the best that can be done is to throw on some filters to increase the compressibility or maybe play with your cc settings a bit more.

Mr. B

[birdy]

@MrBunny
Well my sources don't have any mosquito noise!
So I get them from xvid encoding.

@iago,
You advised me about using a TemporalSmoother in another thread.
I used the one integrated with avisynth2.5 : TemporalSmoother()

I was wondering whats the best TemporalSmoother filter? Should I continue using this one? Also for using it I have to add
ConvertToYUY2() in my avs file.
is it a good idea to go with it?

Anyway the best result I achived till now on the movie I am encoding is with this avs file:

AVISource("F:\Captures\Movies\768X576_Back To The Future 3.avi")
ConvertToYUY2()
Crop(4,84,-0,-84)
UnDot()
TemporalSmoother(2,3)
BilinearResize(576,288)

All comments are welcomed :)

[MrBunny]

@birdy

In terms of temporal filters, I'd recommend you throw temporal or something like it into a search (particularly in the avisynth forum). I'm pretty sure you'll find the answers you're looking for. But of course the best method of learning in this case is experimentation :)

Mr. B

[OUTPinged_]

./me scratches a head

Mosquito noise appears only when you use MPEG quantizer, if your source doesnt have it. If a source is noisy, it will get amplified. Here's what do you need all those temporalsmoothers for.

A simple solution is to use h263 if your encode's average quantizer is higher than 3.5-4, and mod.hq if it isn't.

Always keep in mind that compressibility gain from MPEG quantizer comes at a price and that price is mosquito noise :-)

[MrBunny]

@OUTpinged_

In my experience, mosquito noise (or ringing, or edge noise, or whatever it's called today) still occurs around high contrast edges, even with h263. I think it has more to do with the manner in which xvid compresses rather than the source material. If you try to compress something with h263 at really low bitrates, I'm quite sure you'll run into the same type of effect. While it does occur more often with MPEG quantization, that can be explained by the MPEG quant requiring more bits to depict the same frame.

Personally, I don't use mod HQ as AFAIK, it's not MPEG-4 compliant (you can only have one quant method/matrix per video stream?). And again, to the best of my knowledge, MPEG quant is less compressible than h263, not more. MPEG generally renders more accurately and thus requires more bits (as a generalization).

No offence to you OUTpinged, I'd normally hang on your every word (I owe you greatly for your explanations of the alt cc settings), but this one time, I'm really confused by your post :(

Mr. B

[OUTPinged_]

Personally, I don't use mod HQ as AFAIK, it's not MPEG-4 compliant (you can only have one quant method/matrix per video stream?). And again, to the best of my knowledge, MPEG quant is less compressible than h263, not more. MPEG generally renders more accurately and thus requires more bits (as a generalization).

Hmm, that's very strange. I will make some tests and post results a bit later.

There, it seems it will depend on source, animation will be more compressible with mpeg while regular movies and VHS caps will be opposite.

[iago]

@birdy

It was not me but DJ Bobo who advised you to use TemporalSmoother in another thread ;).

On the contrary, I said I was not much pleased with the results I got with TemporalSmoother in my limited tries.

[Didée]

Originally posted by iago
On the contrary, I said I was not much pleased with the results I got with TemporalSmoother in my limited tries.
Let me jump in to confirm iago's comment.
For my personal taste, temporalsmoother sometimes introduces way too much artefacting on moving objects, especially if skin tones are involved (faces, moving arms, etc.)
I don't know if the exact working principle of temporalsmoother was ever explained somewhere - I never did find something about that.
For me, it looks very much like it performs kind of pixel locking, and it does that in a manner I don't like.
For plain temporal smoothing, I'd go for temporalsoften instead - especially with AviSynth 2.5, where some fine extensions have taken place (secene changes, and mode 2, which seems to be even more hiQ).
Personally, I get addicted to DUST these days - if I only had a 10GHz machine ;)

(Experts, no need to tell me: yes, dust MAY introduce blocking on certain content, I know)

Didée

[OUTPinged_]

Ok, it seems my script for checking filesizes on different quantizers has finished, so here are its results. Vhs = ugly cam capture, very noisy and blurry, film = typical dvdsource, pretty clean with just a slight "film effect" noise. anime=hq and sharp animation clip. I compared them on different quantizers, ME=6 chromaME=1 VHQ=0


quant\mode vhs film anime
2 1,083 1,023 0,928
3 1,001 0,933 0,939
4 1,145 1,079 0,956
5 1,085 1,006 0,951
6 1,182 1,108 1,066
7 1,113 1,054 1,028
8 1,157 1,091 1,056
9 1,131 1,078 1,037
10 1,159 1,102 1,071
11 1,116 1,075 1,045
12 1,137 1,166 1,083
16 1,102 1,078 1,086
20 1,123 1,077 1,108
24 1,124 1,069 1,133


I also did a quick peek on amount of "mosquito noise" there. These are very subjective, and all comments are welcome.

1. vhs clip looked same for h263 and mpeg on low quantizers, and beginning with quant6 mpeg gets better. No mosquito noise was noticed due to heavy regular noise :-)

2. film clip had that mosquito noise. It was noticable with mpeg quantizers under 6, 6+ were noise-free. h263 was producing noise only at quant2, and, to very small extend, at quant3. 4+ were noise-free. At quantizers over 7 mpeg was a clear winner quality-wise.

2a. I also did one check with nomosmooth with settings of 90,10,2,1,3 (radius5+tresh10 temporal smoothing, radius1+tresh3 spatial), and both mpeg and h263 were mosquito-noise free at quant2. The only thing was that mpeg sample contained that "floating walls" quantization artifacts (non - "mosquito" ones) on flat surfaces. It is really hard to notice with quant2, i only mention it to point out that h263 was producing better looking gradients there, not that mpeg was bad.

3. here mpeg was showing clear mosquito noise with quantizer range of 2-6, and at 7+ it got even worse :-( But edges were sharper at higher quantizers, so if you will compress your clip very strongly (40 or less 1/2pass ratio) you would be better with mpeg. h263 was producing more stable picture at quantizers 2-5, was about same as mpeg at 6-8 and later it begun to lose sharpness on edges. Also at 6-8 h263 was producing less "floating walls" artifacts.


So, my advice here would be: while encoding hq anime stuff and never going higher than quant7, i would chose h263. All noisy and/or low bitrate encodes (4-12range) are better to be done with mpeg quantizers and movie dvdrips are dicey.

Also, mod.hq seems to not break compatibility, but ffdshow isnt a best thing to check that. Somebody with dvd/divx standalone should check it and tell.

[cult]

And dont forget the cpu2="ooooxx" prepostprocessing in marc's mpeg2dec3 only for deringing

[digitize]

Personally I like temporalcleaner much better, I don't really get the artifacts that temporalsmoother introduces. It was ported by vlad and works well imo.

[Tommy Carrot]

Don't forget that QPEL amplifies the mosquito noise very badly. In fact, the artifacts can be very visible in encodings with constant quantizer 2. So never use qpel on higher quality encodings than 1cd.

[birdy]

Well I must thank everyone who helped me here, as with your answers and comments to my 2 threads I learned more then all the guides that I was reading over and over!

Tonight I just finished my last 1cd encoding trying to consider all your tips and advices and the result was simply amazing!

I showed my encode to my wife and she was not able to make the difference between my original file and the encoded one :)

The only thing that surprised me was that I never had ANY problem with high motion parts of a movie. The image was as clear as the source! But the very few blocks I could hardly find in the movie where on low motion (VERY LOW MOTION) parts! Specialy on faces and skin parts. The very few Divx movies I ever downloaded were totaly the oposit! Blocks on high motion and good quality at low!

Also a tip from my experience that for sure non of you who helped me would need (but could help newbies like me who would read this thread) is that : I did my 2 pass encodings exactly as the guide available here. But I could never achive a result that could satisfy my needs! But when I did the second pass as external using koepis statsreader the result was really improuved. I was wondering if this is the ultimate auto stats optimizing tool or is there anything that goes furthur?
Anyway for now my life is been a mess! My sleeping time is a mess and My pc is a mess with all the test avi's and avs's and stats!
Yes I have been working really hard for the the past 2 weeks every night till morning! But it finaly payed out and I am going to watch my first nice encoded movie tonight and cleanup the mess on me pc :)

Thanx again to all of you.
Peace.

[iago]

[...]for now my life is been a mess! My sleeping time is a mess and My pc is a mess with all the test avi's and avs's and stats! Yes I have been working really hard for the the past 2 weeks every night till morning![...]@birdy,

Beware man! You seem to be getting detached from real life day by day! :D

iago

[birdy]

lol man,
Only if you knew!
Besides the encoding my real life is just perfect!
Playing Mohaa, GTA, Mafia etc etc,... programing my funcards, chating, posting, checking mail, Formating! reinstalling, tweeking hehe !
On the economic side my wallet is just a mess too when Upgrading!
And when it comes to romance, load a damn beutifull .jpg :)

grrrr someone get me out of here plz :)

[TheUnforgiven]

LOL
that is very common here i guess