One of the universal t-girl traits is a love of photographs. Photos can lie and say you are beautiful. They can make an illusion last. I know it is shallow, but I am no exception and have rarely met a camera I didn’t like. However, photos are also part of my job. Now I’m a not a professional model or a photographer, just a computer scientist and cameras are the eyes of my computer. Our brain has dedicated neurons for recognizing faces, I try to teach computers to do it even better. Under the right lighting, computers can see every pore, every wrinkle, and they never blink. Recently we started capturing faces with macro photography, getting really REALLY close up. At this resolution even the pores have little microwrinkles on top of them. The resulting video represents a texture map of the face with over 16,000 x 16,000 pixels. If you printed this out, it would fill an entire wall. Most of this detail is only visible in the shine of the skin (this is why we powder our noses). Even then, whenever I am photographed at this resolution, I always have a sudden urge to scrub my face.
Computers are also rapidly learning how to identify faces. The best recognition algorithms still require many photographs to account for changes in environment and emotion. This is why Facebook and Google are always asking you to tag all your photographs. As they gather more training photos, it becomes easier to find statistical patterns and recognize your face. All these computer algorithms are designed to classify you by name, age, and gender. However this can be troublesome to some in the transgender community who are trying to transcend these boundaries.
Both computers and our brain look for similar things. Most algorithms start by looking for the t-shaped shadow defining the eyes and nose, then try to match the corners of the brows, eyes, nose, and mouth. The relative proportions of these features, plus the general texture of the face define your facial identity. We can make some generalizations about the difference between masculine and feminine faces. Female faces tend to have higher but less prominent brows, higher cheekbones, fuller, darker lips, and rounder, smaller jaws. Peter Frost has a fascinating blog on the evolution of gender traits in particular differing skin, hair, and eye pigmentation. Research shows that female faces inherently have greater contrast particularly around the lips and eyes. Women also tend to have lighter, paler skin due to lower levels of melanin and hemoglobin (the pigments that make us tan). This also helps explain why we add dark eye liner and red lipstick to accentuate this contrast.
However even for an abstract quantity such as beauty, someone will try to quantify it with a computer. Tommer Leyvand (formerly with Tel-Aviv University, now at Microsoft) conducted a survey to rate the beauty of a set of faces, then wrote software to warp any given image to try to match the ideal face. Unlike the above example, this software does not change the color of the face just the overall geometric proportions. For more technical details you can visit Leyvand’s webpage or watch their YouTube video. There is also a well-written New York Times article for the layperson.
A few years ago, I got to see a talk given by Leyvand’s advisor. In the talk, he noted how they compute different ideals for males and females. He even applied the female aesthetic to his own face though only for comic effect. Of course, unless you plan on having serious surgery on your underlying bone structure, this has limited real-world application. What I find interesting in these images, is that as you try to increase the beauty in each face, you also lose some of the uniqueness that gives it personality. Most of the time, we are trying to cover up our imperfections. However when humans look too perfect, such as in many computer animated films (Polar Express, Toy Story, etc), they start to look artificial and plastic. It is our flaws that make us real.