Digital Makeover?

Dear Reader,

I’m desperately wracking my brain to find some science in this, but I cannot. Nope, reading the article does not help either. Nevertheless, it is in New Scientist (a fact that means nothing on its own, from past experience – they’ve a technology for it’s own sake focus as well, which is fair enough).

Some researchers in Tel Aviv have developed an algorithm that can give a makeover to your digital photographs of human faces. Magazine editors do this by hand all the time, of course, but this algorithm might be able to speed this up, and -for that (I suspect scarily large) number of people who would actually want that kind of thing- allow you to do this to your own photos!

Quoting from Helen Knight’s article:

Software then analysed the images, measuring distances between facial features and ratios such as that between facial width at eye and mouth level, and the thickness of the eyebrows. It compared these with the attractiveness ratings given by the volunteers to create a set of rules, known as the “beauty function”, for assessing whether a face is attractive.

Leyvand has now written a second piece of software that applies this algorithm to a facial image to make adjustments to features so that they more closely obey the rules. It then analyses the results to determine which changes have been successful, and discard any that don’t work. Users can also adjust the severity of the changes.

You can go to the site yourself to look at an example of the results.

I’ve the following questions.

(1) Could someone help me think of an application for this other than the face value one? (Pun only partially intended.) For example, cosmetic surgery, whatever you think of its casual use, is part of a larger set of techniques which are a wonderful thing for victims of accident or disease, helping in the reconstruction of facial features.

Trying hard here, I could imagine that this is related to research in facial recognition technology, useful for automatic identification in security situations, suspect identification, etc. But there’s no mention of this in the article. (But then, it is New Scientist….see my opening remarks)

(2) Is this algorithm working with three or two dimensional information? Picture this: What happens if you change the angle from which the photo is taken? I can imagine this would generate inconsistencies, since the algorithm might make “beauty modifications” which are relevant to that particular angle of the shot. Those modifications (change the slope of an eyebrow here, the height of a cheekbone there) might not match the modifications made for a different angle. This would limit the applicabilty to facial recognition, I’d imagine.

Finally, while I’m being (I admit) po-faced about all this, let me end with:

(3) We might as well go all the way with the ridiculousness here, and ask: Might this work for photographs of pets too?! Just imagine…

-cvj