Why is SixOfNine-HVS matrix asymmetric?

[helge79]

I've just noticed that Didée's SixOfNine-HVS quantization matrix is asymmetric: in the intra matrix, the 7th row 5th column entry is 21 and the 5th row 7th column entry is 20. (And it seems to be like that in all the versions of this matrix I could find at doom9.)

I would have naively thought that XviD compression should treat the horizontal and vertical directions in the same way, and the matrix should be symmetric. Am I missing something, or is my version of the matrix just wrong?.. :confused:

[Didée]

Well, after several hundreds hours of encoding tests, side-by-side-comparisons and double blind tests, using some dozens of different sources, the conclusion was that in that very place, a "21" instead of the theortically correct "20" gives a better ratio of bitrate vs. perceived quality.

In short: it's just an error. :)

(And unimportant, too ... the difference is impossible to spot.)

[helge79]

Thanks for the quick response, Didée. I guess I'll just use "20" from now on. I didn't expect this single number to make much of a difference, but I find it more aesthetically satisfying to work with a symmetric quantization matrix... :)

[JarrettH]

interesting tid bit :p

[Soulhunter]

I find it more aesthetically satisfying to work with a symmetric quantization matrix...

I guess Monk would be proud of you! XD

[rforums]

can any one tell me where i can find SixofNine-HVS Quantization Matrix?

[Jawor]

:search:
http://www.ligh.de/software/qmatrix.zip

[rforums]

thanks