Yep, and it should be remembered that LimitedSharpen needs a good quality source. If your source has artifacts, LS will enhance them as well because it cannot determine between for example MPEG2 macroblocks and detail.

It's also a good idea to use it as a resizer.

I had a big surprise concerning that : I always said also that you must denoise before to sharpen but I tried the opposite and the final mpeg was 10% smaller. And after a closer look, I can tell that LSF().LRemoveDust() is better than LRemoveDust().LSF() !

What do you mean about "using it as a resizer" ? The function use lanczosresize (in a complex way ;)), is the result better than simply using LanczosResize as I do ?

Well, denoising after sharpening will press down the artifacts which have been brought up. I think it might be source-dependant as many say that the sharpener should be the last one in the filter chain. Though if you do strong temporal denoising, the ugly temporal artifacts really blow up in your face. Then again, using Repair with temporal filtering might also be a good idea..
LS does supersampling so it blows up the image, does its magic, then resizes it back. You'll avoid one resizing step if you use the parameters to determine the resolution you would resize to anyway.

This does not answer to the mystery : why the denoiser removes more things after they have been sharpened. Logically, the more a details is sharp, the less the denoiser should consider that as an artifact and it shouldn't remove it. Isn't it ?

It could be the artifacts that appear when you clean the video. Sharp noise will be more like regular noise to the encoder but it will mistake soft noise (dancing blocks you get with strong temporal denoising without motion compensation) for motion and thus creates more motion vectors. Sharpening after denoising will enhance the swimming blocks even more.

This is all just speculation, I have no facts to prove it :?

Okay, guys. I thought I was gonna try out this LimitedSharpen stuff. But do you recommend I use LimitedSharpenFast instead?
And, once again, what filters do I need? I downloaded the LimitedSharpen.avsi and the RemoveGrain 0.9 package. Do I need anything else?
Please, provide links. ;)

See my answer there :
http://www.kvcd.net/forum/viewtopic....654&highlight=

Hello, just want to point out that there is a fairly new update to a mod of LimitedSharpenFaster thats working very good. Its called LSFMod. The speed of LSFMod is little bit slower but the compression results are significant.


LSFmod [v1.9 - Update 2009/10/05]




Comparisons:

Source [583kb]
LimitedSharpenFaster(strength=200) [713kb]
LSFmod(defaults="fast",strength=200) [616kb]

More comparisons in #131 of the thread




Saluts

Thank you. Sounds good, weird as better compression means lesser sharpness.

You followed the thread ? What diff between slow and fast, beside for speed ?

Edit: I love the "show" mode :)

Core differences seem to be higher ss, different sharp mode, more exaustive edgemask and different ways of how to limit overshoot+undershoot (edges, nonedges, etc)

Fast is really good and is all I need for most settings, I love "show" mode too 8)

I saw that slow mode uses a "non linear" sharpening. At the moment I don't see real benefit to use it rather than fast.

I'll compare LSF en LSFmod and report my compression gain.

Looking forward to your real encode comparisons 8)

I was preparing myself to do some encode comparison, and actually I won't.

LSFMod is far less sharpen than original. That explains the size diff.

I did not look at the example on my desktop monitor till now (I was on my laptop the other day). Look at the windwheel on the left: the aisles are less defined with LSFMod. The same way the sky just behind the wheel is more fuzzy (that is a good thing actually as LSF is too sharp in these gradient area).

So, there is no magic :). This is still a good sharpener.

(note: I looked at the example with "slow" too, and that's the same).

Yes i noticed this but for me seems that the limitedsharpen added a little more sharp but also enhaced the noise a lot, maybe thats the reason why size is so different. For me is a trade im willing to spend.

Anyway you can emulate the same lsf result in lsfmod with defaults="old" for still being able to use "show" mode, the size ends up almost equal in both when doing this for what ive tested.

So maybe if you want a little more sharp you can tweak lsfmod for something in between and get still nice size reduction.


Saluts


EDIT: Only by rising the strength seem to get better sharp result than a limitedsharpen and the size is a lot smaller, same source.

Source [583kb]
limitedsharpenfaster(strength=200) [756 kb]
lsfmod(strength=300) [681 kb]

Be carefull. On a live video (= avisynth applied on the fly while Im watching the video), whatever the sharpener (LSFMod or LSF), the strength above 150 leads to awfull result on a 50" screen. Flicking edges and other artefacts that I reduced... lowering the sharpeness of the screen ! This is a nonsense :wink:

Wierd thing is that with same source i got different results than the original examples with limitedsharpen(), maybe he cleaned the source a little before sharpening with both cause also get this artefacts with lsfmod (on the sky).

But is safe goin over 150 when encoding right?

Either you did not understand what I've said above, or you wrote the opposite of what you wanted.

It is NOT safe...

Yes i forgot to add something like "...for all sources?".

So isnt this source dependent? But the comparisons i gave also goes in for example that a safe strength=135 (or even less) in lsfmod with "fast" settings should leave better or equal results in sharpening and a lot better results in compression than a limitedsharpen default strength=100.

Yes, it's source dependant in the way the smaller the resolution is, the bigger the impact of the sharpness.