It is basically a way to do an automatic colour match between two images using one as the source. The great thing about it is the images don't actually have to be the same, which is handy for matching up footage shot at different times of the day automatically.
But it would work even better when the images are pretty close, like two different transfers of the same film.
Most packages just do a median/mean colour match, (Like combustion's colour match tool, you pick the low, the high and the 'average' colour to match to)but in this methodyou get the probability density function for both images (the pdf) and effectively transfer it exactly across to the target image. The result is a pretty damn good colour match without the banding, noise and misconvergence you typically get.
By iterating single dimensional pdf transfers you end up with an extremely close match.

I have the math kicking around somewhere and the siggraph notes, it isn't that hard to implement and gives great results. I've been using it to pull together all of the various sources we have into a cohesive whole, just thought you might find it useful (plus a standalone pdf colour matcher is something that would be very useful, mine is coded into Shake scripts which isn't useful to the majority of people on these forums)

I'll dig out the details and post it over in the tech forum.