KVCD Matrices: Notch versus HVS matrix?
 

Someone willing to compare the HVS matrices with the Notch matrix?
These matrices are very used in MPEG-4, look here:

http://forum.doom9.org/showthread.php?s=&threadid=34493

and the Notch matrix didn't work very well on MPEG-4, according to this:

http://forum.doom9.org/showthread.ph...751#post363751

Since both MPEG-4 and MPEG-2 are DCT based, I'd like to know how the HVS matrices perform in MPEG-2, since I already know how Notch performs in MPEG-4.

Bilu

?? I thought that numbers can't go above 100 !

Don't know much about matrices myself, but the feedback I got from the HVS-Good matrix versus the Notch matrix in MPEG-4 was that Notch got worse quality and bigger filesizes.

I'd like to hear some feedback about the HVS-Good matrix in MPEG-2.


Bilu

Those matrixes are designed for MPEG-4. Not for MPEG-1 or MPEG-2.

-kwag

Here we go:

Nice Post! but not in case of mpeg1 or mpeg2, and Ill tell you why:

As mpeg4 does take MUCH more frames within the intraframe steps, the non-intra values are not mpeg1 or mpeg2 optimal!
If you look closer the non-intra values do cut much more in comparison to the intra values and in mpeg1 or mpeg2 it should be the opposite.

mpeg4 quatisation seen as a whole is different (as I know from doom9.org).
So HVS isn't developed for mpeg1 or mpeg2 when watching the values and their logic.

And in your case its not the best to use Notch in case of mpeg4 as the logic of non-intra cutting is totally different in comparison to HVS.

So maybe "on the first view" those HVS matrixes will give you a big decrease of final filesizes BUT the danger of pumping and rapidely getting worse and worse B and P frames within the I frame interval could be the result!

;-)

But anyhow, thanks for the links, ... interesting!.

PS: I don't know why "mf" an AVS filter designer tries to compare HVS vs. Notch on mpeg4? 8O As he should know bit about it.
http://forum.doom9.org/showthread.ph...751#post363751

And Kwag already answered there.

Thanks for the answer, I really didn't knew that MPEG-1/2 quantizations was that different from MPEG-4, that's why the "designed for MPEG-2" answer wasn't enough for me :roll:


Bilu

Now,...

I've tested the "HVS Best"-Matrix and I have to say it's not so good like the KVCD Notch. I don't know how it is for MPEG-4, but for MPEG-1/2 the KVCD Notch have less blocks than the HVS.

In case of Mpeg4 its already explained in the link given by bilu.
And that it will give you more blocks in case of mpeg1/2 that's shure as explained above (Non-Intra Block).

Inc.

Yes, I know. But I want a result and not a theory and so I've test it and it's clear. :wink: :wink: :wink:

@incredible

How about reversing the matrix, i.e. use intra as non-intra and vice-versa?
It fits with the logic from your post I guess.

Maybe some MPEG-4 matrices could be nice for MPEG-1/2 after all :)

I'm just curious.

Bilu

You can't as that would mean the first value of the Intra 8x8 Block in the left upper corner would be 18 or 19 or 20 depending on what HVS matrix you choose! Thats the first issue, and second if you see the values, in intra-block the values do follow totally different as they do in non-intra.

Watch the 8x8 Block architecture here:

http://www.digitalfaq.com/archives/error.gif

Here's a very good explaining article of KIKA from doom9.de where the basics are explained very well .... to me that was enlightnen! ;-)

Translated by Googles Translation Tools:

http://translate.google.com/translat...language_tools

(well Google seems to translate it not till the end, so you should do a step by step translation using google)

If you want the notch to become even more compressable, just attack the last two collumns in the matrix, like Done in CCE at very low bitrate matrixes.
As we see our KVCDs only on Television wich is 525/625 Lines based and NOT pixelbased, you won't see that big difference. And as we also use in here things like MA (linear blurring unfilter()) or Merge/Luma/Chroma-Blur in our Scripts, the Image will get already in our pre-processing a bit unsharper. So this ULBR Matrix modding won't do much harm to the final quality on your Tv, but you should test it as I changed to Quality priority instead of max less filesize as I mostly do DVD-R encodings now.

Just change the values in the Notch with "99" as seen in here:


In case of KVCD/KSVCD (as it will be stretched horiz. we ONLY modifiy the last row as shown)

Intra

XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.99.99.99

Non-Intra

XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.XX.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.99.99.99



In case of KDVD or KVCD 704x576(480) (more horiz. cutting as it won't be stretched that much afterwards)

Intra

XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.99.99.99

Non-Intra

XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.XX.99.99
XX.XX.XX.XX.XX.99.99.99



But as I said ... for standard TvSets

You got me interested :)

My wife bought me a DVD player for Xmas, and I'm focusing now more on MPEG-2 + TVout than MPEG-4 + PC.

I've been thinking about using an older PC with Linux or FreeBSD using automated scripts to backup to DVD or SVCD, but for what I've seen it's still not worth it.

There are two kinds of solutions around: mjpegtools-based and ffmpeg-based. Mencoder uses FFMpeg, transcode uses mjpegtools, and so on.

mjpegtools is 4 times slower than ffmpeg, but rate control is done right.
ffmpeg's rate control doesn't respect max/min bitrate.

I have a post about it here. Its' a pity though :roll:
http://www.kvcd.net/forum/viewtopic.php?p=61036#61036


Bilu