[XviD-devel] new Chroma_opt option review

suxen_drol suxen_drol at hotmail.com
Sun Feb 16 09:50:49 CET 2003


On Sat, 15 Feb 2003 21:02:03 +0100 elcabesa <elcabesa at inwind.it> wrote:
> i'm testing chroma opt opt adn found that it help not compression, i always 
> get file big than the one made without the option enable.
> i think this problem is due to you try flat very dark area and very light ones 
> in image. 
> i think that this souhld be done only if  there are more than x % of chroma 
> pixel inside a macroblock that are dark or light.
> this is what i mean
> the code is now in cvs  take image and flat chroma in dark area (y<16) and 
> light area( y>235). this make some dark area inside a macroblock flat, and 
> couse you have a non flat MB with a smal flat area  when you do FDCT , this 
> give more coefficent.

yep, further tweaking is neccessary. the 16,235 range can be changed to
increase the chance of chroma flatterning (at the loss of some visual
quality).

> i think that filtering in dark area or light area shgould be done only if a MB 
> has 70% of pixe dark or light.  this way most of the MB is made flat nad you 
> reduce high freq coefficents after FTCT
> nowaday , the result is only a bigger filesize (0.1% i think =) ) couse very 
> dark and verly light area are very few

it's experimental. the idea was invented by mf some time ago, and has
been sitting unused in image.c for months. at the request of some users,
i added a general flag, so that ppl can experiment it without delving
into xvidcore. 

hopefully some types of video compress significantly better with
chroma optimation. if not, then we'll surely remove it.

-- pete; life is like a box of ammo




More information about the XviD-devel mailing list