Question about Using Custom Quantization Matrices

[CubDukat]

One of the people who answered come of my questions about handling 3:2 pulldown suggested that one of the things that might enhance my HD downconversions would be to use a custom matrix.

I ripped a section of "Saving Private Ryan" and copied the matrix DGIndex said it had into CCE SP, and the results were excellent with a segment of the "24" season finale.

My question is this:

--I took the matrix from one specific section where there was a lot of motion (the opening battle sequence). Does the matrix change from scene to scene or is it constant throughout the entire movie?

--Should I use different matrices for different types of programming (one for fast motion, another for sitcoms, etc.) or does "one-size-fit-all"?

It seemed to work all right for the "24" sequence, even though there was less motion than in the actual scene from "SPR."

[vhelp]

I am no expert in this area, but I have done my fair share of study, and to
the extent that I can move further in this area, and even make some of my
own comments on this topic. As such, don't blame me if you don't agree..
because no-one is forcing you

I took the matrix from one specific section where there was a lot of motion (the opening battle sequence). Does the matrix change from scene to scene or is it constant throughout the entire movie?

This varies from DVD to DVD. And, there is no "standardized" ruling which
matrice one uses. In fact, its only been recently, that there has been
some evidense of a change in (shall I say) practice -- matrice optimization.

On some dvd's, they will have one "matrice" throughout. And, on some other
dvd's, there may be several.

fwiw, matrice are a new endeavor. Not that its new, but that either we haven't
seen proof of matrice manipulation till now, or we are just realizing the uses
of manipulating them to our advantage. So, it would seem that the Industry
(if I may say that) and User's here and there, are now staring to realize its
potential and we are now exploring this old (but new) teritory, more and more.

(to my recolection, matrice optimization got its first kick-start "here",
somewhere's around THREE years ago, give-or-take)

** "here" as in, Internet.

Should I use different matrices for different types of programming (one for fast motion, another for sitcoms, etc.) or does "one-size-fit-all"?

The best answer to this question, is a simple one.. Experiment.

I will say this.. its not a good idea for a beginner to embark in this dark
area.. simply because it is not completely understood, and also because its
not for beginners. IMHO, its for those who have been encoding in MPEG for
many years, and especially those who have tasted "advanced" MPEG processing.
I'm not saying that you shouldn't persue this area, but I am saying that
you are jumping inside a (dark) advanced area too soon before you even had a
chance to understand much about MPEG encoding.

It seemed to work all right for the "24" sequence, even though there was less motion than in the actual scene from "SPR."

Matrice optimization requires a special (if not, basic) understanding of MPEG.
That means, processing many videos for MPEG and baring the scars that go with
this, and ultimately forming the neuounces that *cause* you to think for your-
self, in terms of what would bring your MPEG's to better perfection.

In matrice manipulation/optimization, the key to improved video quality, is
knowing how the matrice values effects and exports to the final MPEG in terms
of total image reproduction. This requires prior knowlege in MPEG encoding.

For instance, in your last quote, you basically wanted to know what was best
for most videos. The honest answer, is no one really knows for sure, until
they had a chance to "analize" the source and make projections.. some, based on
theories; others' from on-going endeavors of calculator guess work; while yet,
some even in long-term study via image analisis work; and last, even in a
battery run of trial 'n error encoding scenarios.

Then, over time, a new set of neuounces form, and you have increased your
MPEG video knowledge further. And, the evidense could come in the form of
tools or guides maps, etc.

IMHO, matrice optimization is separate form of art.

Did I mention that I am no expert

Cheers,
-vhelp 3995

[pijetro]

Well, lets' step back and have a look....
Is the footage clean, progressive, inverse telecined etc?
Is the bitrate high enough?
Even at default settings, given clean footage, CCE will give great results..First and foremost.

The fact that you can fart around with custom quantization matrices, lets you get to the nitty gritty of encoding a customized session..
The fact that CCE lets you only do this once at the beginning of the session would point me in the direction of answering "NO" to the original question..

--I took the matrix from one specific section where there was a lot of motion (the opening battle sequence). Does the matrix change from scene to scene or is it constant throughout the entire movie?

Whether it does or not, CCE won't give you that option anyhow.
Although tweaking the matrix might give you an edge, the fact of the matter is, that if the source material is decent, and the bitrate allocated is enough, then the output should be fine...

[manono]

Hi-

Even at default settings, given clean footage, CCE will give great results..First and foremost.

Maybe yes, maybe no. At default settings there is filtering in place that softens the picture. At default settings the Standard Matrix is used that softens the picture even more. The upshot is that at default settings a lot of detail and sharpness is thrown out, crippling your encode, in my opinion. If the source is a very high quality Hi-Def capture, or a top quality DVD, you can do much better by disabling the low-pass filter, and by using a better quantization matrix. He's not talking about making up his own, but using one from a good quality Hollywood DVD, one that will be much better than anything that CCE offers. If you'd bother to check, you'd know that Hollywood DVD Production houses quit using the Standard Matrix a long time ago. The ones they use these days are way better.

Whether it does or not, CCE won't give you that option anyhow.

Sure it does. If you uncheck "Disable Adaptive Q-Matrix Switching", then you'll get the Standard Matrix and (usually) 2 more, the Standard Matrix with the numbers halved (a good one), and the Standard Matrix with the numbers quartered (a VERY good one), and they can change as often as every GOP. You can work this Adaptive Quants with any matrix as a "base", not just the Standard Matrix.

Many Hollywood DVDs these days use the Standard Matrix with AQ, and we enthusiasts can do exactly the same thing. If you don't want to learn about matrices, unchecking that box is a good way to immediately improve the quality of your encodes. You'll have to have a version of CCE more recent than 2.50, though. Also, if you're one of those that likes to get away with only a couple of passes, you'd better let it run another pass or 2 for CCE to get the best matrix optimization, and for it to zero in better on the size you set.

[pijetro]

Sure it does. If you uncheck "Disable Adaptive Q-Matrix Switching", then you'll get the Standard Matrix and (usually) 2 more, the Standard Matrix with the numbers halved (a good one), and the Standard Matrix with the numbers quartered (a VERY good one), and they can change as often as every GOP. You can work this Adaptive Quants with any matrix as a "base", not just the Standard Matrix.

I've gotta look more closely..Of course, there's presets available, but i've never seen the options you point out..
My bad..

[CubDukat]

Well, I just did my first real test with the "Saving Private Ryan" matrix on last night's rerun of the "24" season premiere. The video quality was excellent, but there was another side effect that I hadn't taken into consideration.

The bitrate was quite a bit higher than I expected, It stayed mainly in the 7-8mbps range, even during the darker scenes when it would have dropped to at least 3 1/2, if not lower.

Because of this, the title has gone from fitting on a single-layer DVD to needing a dual-layer DVD--which I do not currently have. So what I'll probably end up doing is burning two versions--a dual-layer version with no further compression and a version that can fit on a regular single-layer DVD. Fortunately DVD Author 2.0 will crunch it down, but it'd be a shame to do it.

Guess the lesson of the day is that maybe that matrix is a little too much, even for a show like "24." At least it is unless I want to make each two-hour block a dual-layer title (which I would want to do, but could not afford to do right now).

I'll keep experimenting with different matrices, though. I'll try ripping a fairly aggressive sequence from "Revenge of the Sith" tomorrow and plugging its matrix into CCE.

[manono]

Hi-

Guess the lesson of the day is that maybe that matrix is a little too much...

Only if you're in OPV Mode will the matrix used effect the file size. Otherwise it's entirely up to the bitrate you set, and the matrix has absolutely nothing to do with it.

[CubDukat]

OPV mode is single-pass VBR, right?

That's what I was using.

I'll have to try it with 2-pass.

[manono]

Yes, OPV= One Pass Variable Bitrate. A decent matrix can easily double the size of the video when compared to the Standard Matrix, when doing a Constant Quality encode for the same quality. And that extra size translates into better detail and sharpness retention.

[CubDukat]

Well, I finished the first part of the multipass reencode of "24," and I'm still impressed with the quality--and even more so with the file size. I might actually be able to squeeze this onto a single-layer DVD+R without more compression.

I did a side-by-side comparison between each segment and found that there's not any noticeable difference between the two files, despite the lower bitrates on the multipass version.

I set the average bitrate at 6318 and the high end at 9341, according to the calculator on this site, and I ended up saving the setting as a preset in CCE SP.