Question about Quantization Matrices

[kschang]

Is there an article explaining how quantization matrices work, and how they affect quality? Or perhaps, how the encoding process works in general?

Was kinda intrigued on how the "KVCD" and whatever matrices they use was able to squash so much space. I've seen the KVCD output, and like guns1inger I'm not too impressed, as it has artifacts all over the place, but for SOME applications it may be suitable.

[mikiem]

There's a lot of articles & info out there regarding video compression, custom matrixes and so on. Also SnellWilcox has a number of great, I guess you'd say small books available for free download. They date back to the mid 90's, but the basics of mpg2 are covered, along with (for me anyway ) valuable info on the analog signal aspects, stuff that's still used/referrenced in mpg2 today.

[manono]

Don't know if this is what you're looking for or not. The matrix discussion begins about half way down in the Intraframe Compression section:

http://www.iis.ee.ic.ac.uk/~frank/surp99/report/kht97/work.html

[kschang]

I *think* I've figured out what they are. Quantization Matrices acts basically as "filters" for the video stream before it's fed to the encoder. The filter can be used to FORCE more data to be the "same" (smoothng, in a way, both physical and temporal) in order to make the encoder more efficient, or vice versa.

KVCD and those "XL length" types use a QM that's extremely softening and lossy to achieve those super-low final bit-rates. That QM filtered out a lot of the content.

Regular MPEG / MPEG2 QM's are a lot more "conservative", so to speak. I'm guessing the original committee did a lot of research and arrived at the current default values as a "best compromise" of sorts.

[manono]

Default values? Take a look at the matrices used in retail DVDs. There's dozens of them. Some are used more often than others, but there's nothing default about a matrix. Now, if you're talking about the Standard matrix used as the default matrix by most MPEG-2 encoders, then yes, I guess it's default, but when making DVD backups, it's about the last matrix I want to use

[vhelp]

TGIF everyone

Also,

Regarding Matrice Effecitiveness ...

There are reasons why using one matrice performs better than another
matrice. But, you have to understand that the source's "lighting"
level plays a part in the matrice's abilities.. and a few other
video attribiutes as well.

In other words, just because one matrice is good in one scene does
not mean it will continue be good throughout the entire source.

Thus, there are various Commercial DVD's that have fluctuing matrices
within them. You only have to research this through several of your
own DVD's to learn this.

IMHO, the most effective matrice (doen't exist) but rather is when
the MPEG encoder encorporates a method to "dynamically" compute the
individual matrice(s) to use during the MPEG encoding phase.

There are many good uses for such a system, but is beyond the scope
of this topic.

-vhelp 3745