"I Wish They All Could Be ..."
02/12/2011
A+
|
a-
Print Friendly and PDF

At 1440 Minutes Hate, Emmanuel Goldstein is making up composite portraits of college age young ladies by state from the first usable straight-on facial pictures he finds on line.

Here are Alaska, Arkansas, and California, respectively. His face recognition software package doesn't work well on blacks, so he is doing them separately. He's including whites, Hispanics, and Asians in the composites. Thus, the California girl looks very anime, like a blonde sorority sister in a Japanese manga comic book about life at UCLA. That's not that surprising when you think about it, since the Japanese comic/animation industry is built on the style of the California-based Walt Disney, with characters being made to look vaguely Japanese (but with round eyes).

Here's his Alabama girl and Arizona girl.

He says, "I used pics from Myspace and color photos from various online college newspapers."

In general, this process has some inevitable biases built in in terms of the source photos. Girls post online flattering pictures of themselves and don't post unflattering pictures. Some of the pictures are photoshopped to look better. More attractive pictures will tend to float to the top of search engines and have more links to them on social networking sites. This cream-rising-to-the-top of the Internet process will advantage populous states, since the first girls you find from California will be higher in percentile terms of all that state's girls than the first you find from Alaska. Thus, the frontrunners for #1 state in attractiveness in this kind of competition would likely be California v. Texas v. Florida.

Finally, the averaging process removes asymmetries and other idiosyncratic flaws.But it can't do much for weight other than to skim the cream. I haven't been to Alaska in a long time, but I was surprised a few years ago by how much chubbier Oregon college girls are than California college girls.

Beyond comparing states, there ought to be a lot of interesting things you could do with traits other than location. For example, what is the composite face of a Taylor Swift fan vs. a Lady Gaga fan among white girls 18-22? Republicans v. Democrats?

Victorian novelists used to go on and on for pages about the shape of the noses of their characters on the theory that facial types correlated with personality. This software lets you look for correlations.

Other surefire winners would be to compositize Facebook pictures from rival colleges such as USC v. UCLA, Harvard v. Yale, Smith v. Mt. Holyoke, Notre Dame v. Miami, UC Santa Barbara v. UC Santa Cruz, Cal State Northridge v. Cal State LA, MIT v. Cal Tech, Tulane v. Vanderbilt, and so forth.

Print Friendly and PDF