"There's only one hard and fast rule in running: sometimes you have to run one hard and fast."

Wednesday, April 20, 2016

Most Beautiful Woman 2016

People Magazine just released their issue declaring Jennifer Aniston as the most beautiful woman of the year. Earlier this year, I chose 16 year-old Russian model Daria Milky:

I thought it'd be a good time to detail my experiments with finding the most attractive female face possible.

Early days

20 years ago, I participated in a study that attempted to find what men found attractive in female faces and discovered that I always chose the most feminine face and was saying my ideal was somewhere past the end of their charts. Wondering if that made me some sort of freak, I discovered that there were some inherent problems with the study, which were never corrected in future studies; for one, "attractive" can either mean sexually attractive or aesthetically pleasing and the two are not the same - people naturally are attracted to the faces of babies, but do not find them sexually attractive (normally). Women tend to prefer female faces that are somewhat more masculine than men do - which led to a problem in studies of my own, which I couldn't correct after I discovered it.

Breeding photos like corn

Access to a rather clever photographic program and a supercomputer allowed me to do an experiment. I took six photos of what I considered attractive but unusual-looking women (4 European models, 1 South American model and 1 Bollywood actress) and attempted to extract features from them. The process I used was akin to inbred-hybrid family selection of low heritability traits in maize: 1) an average of the photos was made 2) each of the originals was changed to a proportion of the average deemed most attractive 3) a new average was created and the process iterated. This allowed for simultaneous selection for multiple features, corrected for partial and incomplete dominance and epistasis and should have eliminated the noise of experimental error (mutation, if you will).

What happened was that the most important factors became immediately fixed and then others could be selected. First, eye shape changed until the vertical measurement became one half the horizontal (beyond here lie monsters); then overall lip dimensions changed similarly. There were 26 geometric standards that I discovered. The first problem that arose was that the sample of six photos I used was too small, but more became unwieldly; the faces I started with led to some unusual characteristics that could not be altered easily - the eyes tended to become noticeably too large and the irises were no longer circular (an error that makes my photos easy to detect).

Because this could be only a local maximum, rather than a globally "best face," I started doing a ridge analysis, playing with color and expression. I quickly discovered a preference for cartoonishly vivid color. Then, working on expression, found that a "mildly pleased surprise" started to appear; this is akin to what models call "smiling with the eyes" and, when exaggerated, with an open mouth becomes "blow-up doll face." I was going to correct back from that when my computer access was halted. I ended up with:

An interesting, but odd face.
This face most closely resembles, in real life model Monique Olsen (not in the original 6):

My privileges were suspended when I showed that I could change faces as follows:

I used this last photo as a Twitter avatar for a year and many people thought it was real. A new, simplified version of the original program I used was released and I started trying to correct the mistakes I had made. The best face I created was:
The features are much softer than before, but I ended up with this awkwardly cropped photo. The face closest to it that I could find was a German lab's attempt to find the most attractive face:

The celebrity in real life that comes closest is Lucy Hale:

So, here's Jennifer Aniston, and here's an image that's 75% her and 25% "ideal face:"

I think it's an improvement.

No comments: