Hacker News new | past | comments | ask | show | jobs | submit login

The total lack of facial hair, even the slightest stubble on either of them has me classifying them both as female. I know they're the same pictures that are manipulated to make me 'believe' one is male the other female, but there is much more to identifying faces than just contrast, and absent other clues I seem to default to 'female'.



The reason there is no stubble is that these photos are composites (averages) of multiple faces. Any small irregularities in the original photos (not just stubble, but also wrinkles, blemishes, and facial asymmetries) are smoothed out through this process.

Incidentally, this is one of the theories behind why facial composites tend to be more attractive than the sum of their parts. Here's an interactive demo:

http://faceresearch.org/demos/average

Also, you can experiment with your own images here:

http://www.morphthing.com/


This is a classic problem of averaging. The average of multiple data sets doesn't retain the average of all the properties of the originals. A simple case of this is averaging two similar images: the local variance will go down because you've blurred out detail and noise that varied between the two images.

Thus, as you average large numbers of images together, the composite you get is not necessarily representative of any of those that went in, and for that matter is not necessarily useful for any purpose at all, given that it didn't retain a lot of the properties of the input images.


Thats a great point. Two things come to mind:

1) Even if your wrong, and this does make a statement about how we perceive real faces, then contrast would only be a really useful indicator in the case of a very average face. Generally speaking it would probably be a very weak variable in our analysis.

2) It seems like they needed to find a way to separate out just the experimental variable in a way to make its amplitude higher than other variables. The only other way I can think of is to find someone who looks pretty much like the image. When looking at it, I don't see anything that makes me think its a composite (or strange, etc...).


Interesting how you get a more feminine face if you average (even if you average only male pictures).


I think you just blew that site up by posting the link here!


Totally agree. The contrast makes no difference as to how I perceive the face. the look the same, except one is more tanned or something. Illusion fail, if you ask me. The faces seem female-shaped and the smoothness of the skin enforces that for me.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: