had a nice little talk with darius(z) yesterday… and i’ve pretty much figured out how to do the bit-distribution for multi-channel images:
let 1..n be the channels.
start w/ compressing each of them with equally distributed bits.
then let i be the (r)mse [(root) mean square error] of the worst channel (max error) and j be (r)mse of the best channel (min error). furthmore let ib and jb be the bits allocated to produce i and j respectively.
then the new bits allocated to channel i are (ib+jb)*j/(i+j) and for channel j (ib+jb)*i/(i+j). we do this until all channels are “reasonably” balanced (as by our error-criterium).
i only hope that it doesn’t oscillate.. °_°

all the best things in life are small, free and useful… stuff for your palm:

• DiddleBug 2.50b11
• Dokusha 1.3
• EasyCalc 1.16
• LyMe 2 (MatLab on your Palm)
• MiddleCaps 2.6
• PocketKanji 0.4.7
• Razor 0.4.1 (c++ game-development framework)
• SC-103PC 1.01 (calculator)
• Vexed 2.0
• X-Master 1.4 (free hackmaster)