[XviD-devel] adaptive quant question

[Marco "elcabesa" Belli]

im' looking at adaptive quant/lumi_mask code

if i'm right you calc 
image average luminance
macrobloc average luminance 
and then raise quantizer for very dark and very bright area
then normalize each difference between consecutive quantizer due to mpeg 4 
restriction

am i  wrong?


lower quantiuzer (1,2 ) mean an image full of detail, high quantizer (30,31) 
image where only low frequency detail are represented.

why not bootst quantizer for mblock with lot's of DFT coefficient. i think 
this can be doing somethink like lumi mask, using variance, doing lumi mask 
in dft trasformed image.. i don't precisely know.. but i think this would 
give more bit at mblock with lot's of detail , like object edge , and saving 
some bit for plain block