|
Not alot to show for the time I spent doing this. There are lots of flaws so far...
The system I'm using (FACS) is one developed by psychologists in the 1970s to describe facial expression. Any facial expression can be described as a value for one or more of about 50 action units. It poses some problems for me. One, it costs quite a bit, and I've not been able to find any libraries that have it. It comes on CDROM as a pdf, so if anyone would know how it might accidentally fall off the back of a digital truck... It also uses a scoring system that is a little annoying. Each action unit can be scored from A-E, or 1-5 in something a little easier for my database to digest (all unscored action units presumably being zero). My database, interactive face animation, and slider will all be much happier with 0-255 though. Should I ditch their coarsely-grained scoring system for the byte? Should I ever need to reconvert back to the traditional scoring system of A-E, that's trivial, but once my 0-255 slider values are lost, they're gone for good. Finally, there are some action units that just don't make sense. Some, like "blink" aren't really appropriate for still images, though they still have merit for video. Others make little sense, given my posable stick figure plans... "head left, right, up and down" are much better suited for my own joint XYZ/angle scheme. Then there are the ones I want to normalize a bit, why have "eyes left" and "eyes right", when I can have "eyes horizontal" where 0 is max left and 255 is max right? And finally, some just seem to be missing. There is a single "tongue protrude" action unit, when I think I'll need to invent a few more. "Liplicking" and "tongue curled upwards" come to mind, maybe more. And I'm starting if I should throw a "pupil dilation" in there just for good measure, too...
|