@ CheronAph
I'll post it tonight because I'm at the office right now.

@incredible

A resizer sharpen only the edges of the pictures, not the plan area. IN fact, we should say that Bicubicresize soften the edges that are normally sharpened by bicubic interpolation, where Lanczos doesn't (I guess, not sure). Because it's a mathematical reality that applying bicubic equations leads to sharpening the edges.

If you want you have a version of unfiltered bicubicresize (that means, a bicubic that do not do this softening) in the library bicublinresize there :

http://ziquash.chez.tiscali.fr/

Well, I know the "bicubic" algorithms from Photoshop. There its explained that a bicubic algorithm will do an precise "interpolated" resize in comparison to a just "pixel cropped/moved/doubled" resize.

Or does Avisynth got its own definitions

And "Lanczos" .... hmmmm
I did a lot of tests (also with the resizers in the link you posted, thanks ) and by using lanczos instead of bicubic I had to set the values of the following cleaners higher (to handle the plain parts in the movie!) and if not set higher the picture got still noise, cause of the paradoxing-Effect

But I'll test again

Hey Phil, doing Forum Jobs at Work like me?

No no, avisynth have the same definition. But "interpolating" means "introducing an error'". After this, all depend on what you decide to do with this error : drop it completly, or try to counterbalance it with the next pixel you have to interpolate (kind of "error diffusion" for make it short). When you arrive on an edge, the algorithm do not have "next pixel" to use to counterbalance the error acumulated from the previous ones. That is where the problems occur.

This are old memories of my studies, so I'm not very sure of anything

For your last point, that sure that Lanczos introduces some artefacts, but my point is "Lanczos > Asharp+Bicubic".
In your case, as you do not use Asharp, that is a different situation. I agree with you.

The problem now : what are you doing if you find the picture too soft

In my last post I meant the option "bicubicresize & Asharp(1,2)", sorry for my english.

First bicubic then asharp gave me a well adaptive sharpened, cleaned and still detailed (espec. in hairs, skins etc.) image on tv in comparison to Lanczos.

The pics above are in deed handled by using only bicubic in combination with mergeLuma/Chroma-blur without asharp(). Ok, its looks a little softer but so this 180 minutes movie @ 480x576 fits on one CD-R80 and it appears a very very lot more sharper in comparison to a 352x288 encode, even sharper as a 352x576.
If I got an 120-140min, I remove the Merge... Lines cause of more CQ and shure by doing this it looks even sharper

But testing continues .... as there will release new filters and techniques

Ok, if I understand well your point :

- Bicubic then asharp = nice image, nice details
- Asharp then Bicubic = image with noises

I do not really see where can be the difference (except if you put some filters between the two lines). But I never tested the first order.

i did (thousands times)

- Bicubic then asharp = more size, less details!

- Asharp then Bicubic = less size, more details and
more contrast !

- Bicubic without Asharp = less size, loss details!

- Lanczos without Asharp = more size(big), more details and
strange artefacts in the edges!



<edited>
oops...i forgot:
my taste...
encrease the resolution to 702x480,
then you don't need asharp or lanczos.
use only bicubic!

And more noise because you sharpen a noisy picture and asharp doesn't do any difference between a "good" and a "bad" pixel.

If someone can do me a sample of mpeg where these artefacts are really annoying, I take it.

Phil,

"and more noise because you sharpen a noisy picture..."
yeah,can be but my sources are clean dvds,then i don't got this problems.


the "strange artefacts" using lanczos seems "moving ants"
in the edges on low resolutions like 480x480 or less.

do you want pictures or little samples with and without lanczos?

Wow it seems that this thread is getting a little confused .... but even mor interesting.

Well not definitely, that depends upon the threshold-value in Aharp() I think.

You forgot to mention the cleaner/denoiser
I meat that if you first use Lanczos (which sharpens the movie and in my opinion it also sharpens to much other things than only the edges) and if after a laczos there will be a temporal or spatial cleaner the work will be a little paradox.

So I meant:

- "bicubic" then "cleaner" then "a little asharp (if really needed and with threshold set only to handle the edges)" = nice image, nice details, sharp

- "lanczos" then cleaner = almost the same noise when using the same values in the cleaner-filter, so we have to rise up the values = maybe less quality in hairs & skins cause of more agressive cleaning values.

- "bicubic" then "cleaner" = nice image, enough details on Tv AND more CQ!

- "bicubic" then "cleaner" then a bit "mergel-uma/chroma-blur" to break the edges = nice image, still enough details on full vertical sizes like 576px (as seen in my pics above) just a touch softer on Tv but a rised CQ when predictioning for example long movies like my 180min sample above

On the other hand, if we use a sharpening with values of asharp(1,4) before resizing, the sharpen advantage can be killed again by the following downscaling.
Ok, maybe we will receive in relation an image condition as before resizing?!
Like a compensation? I don't know, cause every source is different.

I can proof this, well artefacts or not but the picture looks a bit overtreated when watching on TV .
I mean if a picture is sharpened to much it will look very good on a pc, well amazing, but a Tv and its interlacing can produce "flickering" when used to much sharpened streams.

And I think we should keep in mind that if our source is for example an 720x576 anamorph DVD/m2v stream we resize "down" to our desired resolution, so its not really necessary to do a sharpen before or after resizing cause we will preserve enough sharpeness when scaling down from a bigger detailed source. The case will be different if we have to "scale up" like Dvix/Xvid sources.

Well IMHO everything depends upon the quality of the source, of someones taste and so on. All this all is only my opinion based upon my experiences and testings but shurely even for me not definitive and thats what I like in here.. the discussion

Yeah ! I forgot this essential point and was thinking only "theorically" . But you're right

Yes I do, if you can capture that !

I do not forgot, and you can see that I was wondering about that point

It is not Lanczos that sharpens, but Bicubic that softens ! :banghead:

I often mention a clip I have where the hero has a 2-day beard. Lanczos is the only resizer that does not litteraly shave the actor.
Bicubic isn't only a resizer, it's a Mach 3 razor

Less paradoxal than doing that after an asharp, itself done after a softening bicubic. But we can discuss it all the night

Not enought details for me in fact. So I had to use a sharpening trick. That is my goal in using Lanczos. Teh best will be to find the image produced by bicubic sharpen enought. Because Lanczos makes my CQ drop by several points. I know that.

You're right. That's an other reason to not making it this way.

Find the brakes and stop the train ! I found were is your/mine problem : I have a 100 Hz (the double of the normal PAL frequency) TV set. No flickering at all, like a PC monitor using high refresh frequency.

That perhaps were all is our difference

I never do upscaling.

yes friends,you are very clear in your explanations...thanks!
or we have eagles eyes or it depend of the source, don't?

in the end is just a little differences!

.." that depends upon the threshold-value in Aharp() I think."
.."It is not Lanczos that sharpens, but Bicubic that softens ! "
..."I never do upscaling."
yeah!


"... litteraly shave the actor. "

@ Dialhot

You lucky one, exact thats the point! That will be my next investment a 100Hz living room goal! Maybe Christmas

@ Jorel
right!

You guys are going way off topic , I still haven's solved that stupid problem. Can't i download the right script for the job somewhere?

The answer to your question was in the second post of this tread
It seems that your problem isn't to DL the script, but to know what to do with it. For that you can go on the home page of KVCD.net and find guides on the right panel. They'll give you all informations needed to do a KVCD.

Well, where is it?

__________________
¨¨°º©©º°¨¨°º©CHERONAPH©º°¨¨°º©©º°¨¨

@ incredible

I don´t get it, I used your script,
480x576 mpeg1
bitrates 64 ~ 3000
lenght 98 minutes
audio 128
CQ 63,160

__________________
¨¨°º©©º°¨¨°º©CHERONAPH©º°¨¨°º©©º°¨¨

Here´s the script I used, have I done it wrong,

LoadPlugin("E:\-=[ KVCD ]=-\Filters\Avisynth_2.5\MPEG2Dec3.dll")
LoadPlugin("E:\-=[ KVCD ]=-\Filters\Avisynth_2.5\GripFit_YV12.dll")
LoadPlugin("E:\-=[ KVCD ]=-\Filters\Avisynth_2.5\STMedianFilter.dll")
LoadPlugin("E:\-=[ KVCD ]=-\Filters\Avisynth_2.5\undot.dll")
LoadPlugin("E:\-=[ KVCD ]=-\Filters\Avisynth_2.5\VSFilter.dll")

Mpeg2Source("D:\DVDtoKVCD\vts_01.d2v")

undot()
Limiter()
GripCrop(480, 576, overscan=3)
TextSub("D:\DVDtoKVCD\vts_01.srt")
GripSize(resizer="BicubicResize")
STMedianFilter(8, 32, 0, 0 )
MergeChroma(blur(1.50))
MergeLuma(blur(0.1))
GripBorders()

__________________
¨¨°º©©º°¨¨°º©CHERONAPH©º°¨¨°º©©º°¨¨

I'm not sure what the topic is here but one little tip is to put the dll's into the PLUGINS directory where Avisynth is installed and then you don't need the 'LoadPlugin' lines in your script

I´ll do that!

__________________
¨¨°º©©º°¨¨°º©CHERONAPH©º°¨¨°º©©º°¨¨