XviD QM Collection

[mod]

Hi all. I've searched in the forum but I didn't find nothing similar, so..

1- I'm making some testing with XviD and I'm looking for something like a webpage with a collection of XviD's quantization matrices or something like that.

2- (where) can I find a detailed description of QM developing? I've tried with google but all I found was too "easy". I'm looking for a more "DCT applied to video" approach :)

3- What about Wavelet Packets instead of DCT? (This is the funny question)

If anyone knows plz post it here, thanks

[boombastic]

I asked for something similar here:

http://forum.doom9.org/showthread.php?t=100118

[mod]

Thanks, when I'll have the time maybe I'll found some interesting matrix following the links there.. :)
Any info about visual relation between DCT coeff position and wavelenght of visible light? I know that top left corner is low and the bottom right is high frequency, but do you know if there's also a sort of band decomposition?
I hope this isn't a too technical question..
I'm looking for some reference on the web..
Thanks for the links :)

[Kopernikus]

The DCT Coeff has nothing to do with the wavelength of light.

It represents an spatial frequency.

The visibility of distortion as a function of spatial frequency is named the Contrast Sensitivity Function (CSF). Try Google with CSF for more Infos.

[unskinnyboy]

This (http://forum.doom9.org/showthread.php?t=90954) thread might help for Qn #1.

[mod]

The DCT Coeff has nothing to do with the wavelength of light.

It represents an spatial frequency

I've read about the CSF, that isn't what I was talking about. When I said wavelength I was meaning that human eye is more sensitive to green-yellow components than to others. I know the DCT, but I don't know what the DCT coeffs represent the spatial frequency of in a video frame?
I try to explain better: if you choose to use low "low frequency" values for the DCT parameters, than you get less resolution in the low frequences? This means less resoution for the high wavelengths?

Is it possible to assign really few bits to low and high freqs? I' making some testing and I'm pretty interested in this.
Thanks for the answers, I'm sorry if I'm not clear but this isn't my language :)

[Kopernikus]

The DCT isnt about wavelengths. The wavelength of the emitted light of the monitor is totally uninteresting from a compression point of view. The color that is percepted by the human eye is created as a mixture of three base colors by the display. But that isnt important in this context.

Most video codecs work with a special colorspace called YUV. There the "color" is composed as a mixture of a luminance signal (Y) and to chrominance signals (UV).

Because the human eye is more sensitive to changes in luminance, the crominance signals can be compressed stronger without impact on the percepted quality.

The entries of the QMs specify a sort of granularity. The lower the entry, the less rounding errors you get when reconstructing this frequency component of the frame. Low Entries = precise reconstruction, bad compression, High Entries = bad reconstruction, good compression.

So because of the lower sensitivity of the human perception to higher frequencies, the entries correspnding to higher frequencies in the QM can be chosen higher, as this increases compression.

I hope I understood your question as English isn't my native language either :)

[mod]

The color that is percepted by the human eye is created as a mixture of three base colors by the display.

I know.

Most video codecs work with a special colorspace called YUV. There the "color" is composed as a mixture of a luminance signal (Y) and to chrominance signals (UV).

No doubt.

The entries of the QMs specify a sort of granularity. The lower the entry, the less rounding errors you get when reconstructing this frequency component of the frame. Low Entries = precise reconstruction, bad compression, High Entries = bad reconstruction, good compression.

This is the answer i was waiting for :)

So because of the lower sensitivity of the human perception to higher frequencies, the entries correspnding to higher frequencies in the QM can be chosen higher, as this increases compression.

High frequency <-> low wavelength

OK, I was trying to say the same thing, so you've confirmed me, there has been a misunderstanding..

Nice help, thanks :)