January 15, 2007 9:57 AM   Subscribe

The Morphable Face Model "captures the variations of 3D shape and texture that occur among human faces. It represents each face by a set of model coefficients, and generates new, natural-looking faces from any novel set of coefficients, which is useful in a wide range of applications in computer vision and computer graphics." Amazing/terrifying tech from Herr Prof. Dr. Volker Blanz.
posted by gwint (38 comments total) 11 users marked this as a favorite
Wow! That's some very nice work. The frown stuff seemed least polished—they certainly managed to automate some sort of negative expression, there, but it looked more like Oh Shit Fresh Lemons—but that's a small quibble.
posted by cortex at 10:08 AM on January 15, 2007

Officer, that's not my face on that security video.
posted by anthill at 10:11 AM on January 15, 2007

Making fake porn pics should get a lot better with this.

Aside from the snark, that's some pretty impressive stuff.
posted by effwerd at 10:17 AM on January 15, 2007

Okay, I was all set to be snarky and ask, "All the variations of human faces?" and link to unusual-looking people like Michael Berryman, Rondo Hatton and Joseph Merrick, but that video was just amazing. The way the software can extrapolate from a still image... ladies and gentlemen, I believe we have crossed the Uncanny Valley and are safely on the other side.
posted by Faint of Butt at 10:26 AM on January 15, 2007

Scary cool. Dead people will be starring in new movies any day now, or, even scarier, combinations of real people. You want someone for the part who's a little Angelina Jolie and a little Kirsten Dunst, just morph one up. The best part is you won't have to pay them points or deal with their inane demands for deluxe mega trailers and on-set personal chefs.
posted by doctor_negative at 10:29 AM on January 15, 2007

Oh my god that is beautiful. I wonder if it will be released as something for us all to play with...
posted by twine42 at 10:34 AM on January 15, 2007

doctor_negative, that'd work for me. I'd like to see what you'd get with, say, 40% Julianne Moore + 60% Mae West.

Too bad Star Trek is off the air -- they could come up with aliens whose faces actually move in response to emotional cues.
posted by pax digita at 10:41 AM on January 15, 2007

The only issue with this technology is that it isn't that useful for animation where one needs to exhibit believable facial expressions -- the straight interpolation models lead to stiff or unnatural looking results. But given more study, it is likely to happen.
posted by bhouston at 10:45 AM on January 15, 2007

I look disturbingly like the average face. I don't really know if that's a good or a bad thing.
posted by Alex404 at 11:03 AM on January 15, 2007

Watch this to the end. The 3-D Mona Lisa is a heart-stopping sight.
posted by Faze at 11:33 AM on January 15, 2007

All I can think of is the end of The Running Man where Killian (Richard Dawson) stages the death of Richards (Arnold Schwarzenegger) by Captain Freedom (Jesse Ventura) using computer-generated images.
posted by Muddler at 11:34 AM on January 15, 2007 [1 favorite]

Alex404, probably a good thing.
posted by forblaga at 11:36 AM on January 15, 2007

Awww...captain freedom link went down. Try here.
posted by Muddler at 11:36 AM on January 15, 2007

Thats the kind of tech. they will soon be using in passports etc.
posted by mightyb at 11:37 AM on January 15, 2007

Mona's got quite the honker. Thanks goodness for 2D.
posted by blue_beetle at 11:59 AM on January 15, 2007

Kinda reminds me of the "scramble suits" in A Scanner Darkly.
posted by skullbee at 12:02 PM on January 15, 2007

Still, it's an awful lot of work to go to just to get Tom Hanks to sign onto the Turner and Hooch sequel.
posted by maryh at 12:14 PM on January 15, 2007

wow, this is great. Where does one find such software? it would certainly help a paparazzi who wanted to put a frown or scowl on some movie star talking to their spouse. Or, if they needed pics for "look how fat/skinny that starlet is..." I no longer believe anything i see.
posted by Blingo at 12:21 PM on January 15, 2007

Doctor_Negative: and if you are rich, powerful and sad you could also generate Kirsten Dunst and Angelina Jolie (I might suggest Jessica Alba for a threesome) for your private movie..

No, seriously, I really want a Spore-like game with this technology in.
posted by darkripper at 12:28 PM on January 15, 2007

OMG. Mindblowing. We're never going to be able to assume we're seeing anything real again.
posted by five fresh fish at 12:28 PM on January 15, 2007

This post reminds me of this (although the .mov link seems to have gone dead which is a shame as it was a great clip.)
posted by Rhomboid at 12:33 PM on January 15, 2007

We're never going to be able to assume we're seeing anything real again.

For the record, there's nothing mindblowing about the resultant models' verisimilitude—current tech is such that we can reasonable create photorealistic fullmotion video of any given person we'd like, and the tech on display doesn't handle the degree of idiomatic detail required to make a performance (rather than a constrained demo animation) believable.

The neat trick is just the automation of the initial construction/mapping of an acceptable model. If you want to stop believing what your eyes, you're kind of late to the party.
posted by cortex at 12:36 PM on January 15, 2007

The reason why average of many faces are more beautiful that individual faces is partly because of the increased symmetry. Most individuals do not have fully symmetrical features, although if you average together a lot of faces, the average is very symmetrical. Asymmetrical features is supposed to be indicative of challenges faced by the organism during development/growth -- challenges which may be indicative themselves of non-beneficial genetics. Thus symmetrical features is taken as a indicator of a good match between genes and environment.
posted by bhouston at 1:07 PM on January 15, 2007

Thus symmetrical features is taken as a indicator of a good match between genes and environment.

It probably also accounts for the greater relative attractiveness of persons of mixed racial descent. Simply averaging out characteristics is apparently sufficient to make a face more attractive.

This is excellent, and will especially useful in gaming. It would allow each character in a game (player or non-player) to be given its own unique face, represented easily by a small set of numbers. Modify it a bit beyond the human norms, and you can generate unique elvish faces, Narn faces, etc. (However you will likely have to store most of them - constantly generating a complete set of new people every time one stops off in the same game town, say, or plays a game on a new computer, would be uncannily disturbing.)

Still, it's an awful lot of work to go to just to get Tom Hanks to sign onto the Turner and Hooch sequel.

This made me think of the possibilities of adding non-human facial characteristics into the mix: Turner + Hooch = what? A kind of jowly weredog, maybe, like John Goodman's character in Monsters Inc. Adding the features of a given animal to random human faces would be an easy way to generate aliens or fantasy creatures.
posted by aeschenkarnos at 1:50 PM on January 15, 2007

I believe the video is from their SIGGRAPH'99 presentation. It is very good work, and shows how useful 3D scanner data can be in modeling. The mathematical methods they use are quite straightforward and well-known, Principal Component Analysis for model generation and Bayesian Maximum A Posteriori estimation for model fitting. FaceGen is the best-known commercial fruit of this line of research. (I wish it had been available when I started working on my master's thesis in 2001...)
posted by ikalliom at 2:34 PM on January 15, 2007

By way of comparison, check out this video on animating Brando's head for the recent Superman re-do. That movie came out very recently, but this technique seems like quite an advancement.

Very impressive. And the issues it raises are very thought-provoking. It seems inevitable that this will be used to map celebrities, animated characters, etc, onto motion captures. Imagine if in doing a movie, you signed away the rights for the production company to use your likeness for "map and cap" in sequels.
posted by adamrice at 2:37 PM on January 15, 2007

Great find, very cool -- thanks, gwint!
posted by Asparagirl at 2:40 PM on January 15, 2007

Very cool. I love watching morphing. something mesmerising about it.

MeFite stickycarpet invented morphing.
posted by nickyskye at 2:45 PM on January 15, 2007

But I don't think it's ever been done with this ability to fake identifiable people in real time.

You could literally slap together a real-time video man-in-the-middle attack in which a prisoners' video testimony is subtley changed to present different emotions and tone of voice, without changing the actual head/mouth movements or words. Witnesses to the actual event would have a hell of a time identifying the video as inauthentic: so much of it would be exactly as they remember.

Cool and scary beans.
posted by five fresh fish at 3:13 PM on January 15, 2007

Maybe not a morph?
posted by nickyskye at 4:21 PM on January 15, 2007

Michael Chrichton predicted this one a long time ago. Sci-fi becomes reality (well, almost).
posted by zardoz at 4:35 PM on January 15, 2007

This kind of technology is already used in some games (even the sliders). The Movies, for example (even if I'm not sure if it use just one head model and one texture - maybe it uses one texture for men and another for women). However the models are very low-res. Oblivion uses something like this too, with better results.
posted by darkripper at 4:50 PM on January 15, 2007

I was at a video game conference some years ago where I volunteered to have my picture taken for a FaceGen demo. It was almost frightening to see how one's face can be "masculinized" or "feminized". It's really mind blowing software.

But on a lighter note, when will we see this used where it is needed the most, for Brian Peppers?
posted by Tube at 7:15 PM on January 15, 2007

No black face?
posted by Poagao at 11:55 PM on January 15, 2007

This makes me want to write a science fiction book about a group called "The Discerners" who have been trained since childhood to be able to pick out the difference between computer generated and real human beings. It would track the foundation of The Department of Discerners who are initially used to ward off the use of faked confessions in court, serving as expert witnesses when video evidence is presented. Stories detail their training, the constant battle to develop the human mind such that it can capture the increasing subtleties of photo realistic faces that mark them as such. "A particular face morphing program has this hiccup, or this software has a particular artifact when emulating smiling/frowning."

Eventually, The Discerners will be called upon to participate in separating out plots to bring down world government's that are incited by an organization that heavily uses face emulation techniques to put out viral video on teh internets of the nation's leaders doing acts in flagrant violation of the law/morality/ethics/etc. Oh, wait... errmm... maybe I'll have to come up with some other world threatening plot.

Some how the story will carry in to the future when holographic technology allows people to change their faces, and the Department of Discerners is subsumed into a worldwide policing organization to help capture criminals who are trying to hide their identities.

The last Discerner will be killed by someone using technology that has at last perfectly emulated the human face and its range of expression.

Now adays you can't trust your eyes, it's true... but I like the idea that once you've been exposed to what the differences between real/unreal are, you can learn to pick it out in other things. For example, I can tell photoshoppery on the news racks after studying the alterations that are made as shown by links from a MeFi post a while back. Looking at these videos, one can see that the face isn't perfectly imitated... Tom Hank's nose isn't quite perfect in the middle example. But you'd have to be familiar with the capabilities of the programs, and have a very sharp eye to capture it, and maybe even then you wouldn't get it as an ordinary, casual observer.

But the Discerners would... I will have to start training a group for the good of humanity.
posted by Mister Cheese at 12:32 AM on January 16, 2007

The really amazing part of this is the creation of a 3D model from a single 2D image.

If you are interested in this sort of thing, download the Facegen demo, which lets you experiment with a similar technology. Be prepared to waste hours.
posted by Enron Hubbard at 5:37 AM on January 16, 2007


This is the sort of stuff my lab plays with all day.
posted by dmd at 7:06 AM on January 16, 2007

Wholly unfair that Chrichton's nightmare-inducing "Looker" has potential to become a reality before I get my flying car.
posted by squasha at 7:56 AM on January 16, 2007

« Older Drug Tax Stamps   |   International Dim Sum Directory Newer »

This thread has been archived and is closed to new comments