Although this is probably old news to y'all, a DARPA/NIST conference (public info, don't worry) from about a year ago discussed the current foci of biometric IDing. Among their findings were that infrared recognition is much more accurate (approx. 85% fewer false reads) than visible spectra, for some reason I couldn't follow.
They admitted that no computer modeling software could recognize human features as readily as humans can, and that a major reason for the difficulties experienced by recog software is that it is, pretty much universally, designed by technicians and programmers -- not artists.
They've actually gone out and hired caricaturists, whose ability to quickly read a face and translate it into a readily identifiable facsimile (even an exaggerated one) is, of course, their meal ticket. They're looking at what features caricaturists exaggerate, and which ones they ignore, as future guidelines for more accurate feature processing.
From where I sit, on the indy video side of the razor wire, there's a pretty large knowledge gap between the creators, the would-be purveyors, and the audience. Y'all have the techie wherewithal to know how you want your content delivered, but very few people I know in the biz have devoted enough time to figuring out how to provide it in a format that satisfies their dubiously artistic integrity.
I know of a couple of local production teams who are trying to do this, but they're driven entirely by profit, which ain't even remotely close to a viable business plan. Everyone around here (DC) freelances for documentary work, and if anything gets posted, it's usually a trailer or demo of their talents.
The true gonzos out there (we call 'em Mathers, in his honor) are few and far between... and even ol' Evan's online limitations are, well, evident.
Slashdot Mirror
Although this is probably old news to y'all, a DARPA/NIST conference (public info, don't worry) from about a year ago discussed the current foci of biometric IDing. Among their findings were that infrared recognition is much more accurate (approx. 85% fewer false reads) than visible spectra, for some reason I couldn't follow.
They admitted that no computer modeling software could recognize human features as readily as humans can, and that a major reason for the difficulties experienced by recog software is that it is, pretty much universally, designed by technicians and programmers -- not artists.
They've actually gone out and hired caricaturists, whose ability to quickly read a face and translate it into a readily identifiable facsimile (even an exaggerated one) is, of course, their meal ticket. They're looking at what features caricaturists exaggerate, and which ones they ignore, as future guidelines for more accurate feature processing.
From where I sit, on the indy video side of the razor wire, there's a pretty large knowledge gap between the creators, the would-be purveyors, and the audience. Y'all have the techie wherewithal to know how you want your content delivered, but very few people I know in the biz have devoted enough time to figuring out how to provide it in a format that satisfies their dubiously artistic integrity. I know of a couple of local production teams who are trying to do this, but they're driven entirely by profit, which ain't even remotely close to a viable business plan. Everyone around here (DC) freelances for documentary work, and if anything gets posted, it's usually a trailer or demo of their talents. The true gonzos out there (we call 'em Mathers, in his honor) are few and far between ... and even ol' Evan's online limitations are, well, evident.