Korean Beauty Contestants – Face Morphing

davvid 13 years ago |

I wonder whether this analysis supports the assertion that these were photoshopped?

http://kotaku.com/blame-photoshop-for-koreas-beauty-queen-cl...

msie 13 years ago | |

I wish this comment stood out more but HN has eliminated displaying of points.

lmm 13 years ago | | |

Huh? Points haven't been displayed on comments for at least 18 months.

marshray 13 years ago | | |

Yes, that was his point. At some point in the past, points were displayed on comments. But, as you pointed out, now they are not.

joshguthrie 13 years ago | | |

That's a nice point about points.

thebadplus 13 years ago | |

surgery first, then makeup, then lighting, then photoshop

CamperBob2 13 years ago | |

I would think there would be even fewer "eigenfaces" if someone had Photoshopped them to look alike.

marshray 13 years ago | | |

Some of the photos were taken at slightly different angles. Perhaps that accounts for some eigenfaces.

davvid 13 years ago | | |

That would make sense. Eigenfaces are notoriously bad about translation/rotation invariance.

onemorepassword 13 years ago |

It's worrying how this keeps getting upvoted despite having been debunked before it even hit HN, and the top comments on both HN and the original post pointing to the article that debunks it.

This is one scenario in which HN could really use downvotes.

PavlovsCat 13 years ago | |

For me it doesn't really make a huge difference if plastic surgery or photoshop are the culprit. So these women don't actually all look like clones, they're just presented as clones; great.

hkmurakami 13 years ago | | |

That's actually a really interesting point about the cultural backdrop of the photos being produced.

PavlovsCat 13 years ago | | |

I think many small changes - each of them incremental, and rather innocent in and of itself - to maximize "market appeal" (where the market is some kind of faceless statistical blob), can also lead to this. If you don't know what you're doing and why you're doing it, and/or if it's not something you really can believe in, it will show, in some form or another.

Sure, I can't blame anyone for not believing in their job of photoshopping fashion pics.. but what I find much more unsettling is that young people do get influenced by this stuff, wether they know/want it or not, and wether the people pushing the stuff believe in it themselves or not. But it's not just fashion, our societies are half-assing a lot of things on a massive scale. Too many hustlers, too little pride in work :(

anywhichway 13 years ago | |

I thought the whole point of the post was the analyses and methods and that the outcome was of little importance, but was only there to make the subject more interesting, gripping, and relevant.

just2n 13 years ago |

All I see is lots of makeup. Might it more accurately be called a Korean Face Art Contest?

But really, if we're trying to compare physical beauty, wouldn't the contest require 0 modification of it? This would likely include banning makeup, any form of plastic surgery, unnatural hair modifications, and clothing that is overly supportive. Is there even such a thing? A cursory Google search suggests no. Even so called "natural beauty pageants" apparently permit caked on makeup, fake hair, fake teeth (not just orthodontics), etc. I don't see a point to these beauty contests, in that case. They say nothing about the person at all.

tsotha 13 years ago | |

Look here:

http://i.imgur.com/7kdgJ32.jpg

Particularly at #18, where it's most obvious. These pictures have been photoshopped to give the ladies the elfin look that's popular right now in Korea.

leephillips 13 years ago | | |

Some of the real faces have a lot of character - and they were all turned mindless-looking by the photoshopping.

just2n 13 years ago | | |

I would argue the same is, or rather can be, true of makeup, in general.

therobot24 13 years ago |

neat, but eigenfaces is really overplayed and outdated, there's lots of better face recognition algorithms that can handle a bit of distortion/deformation that really plagues principal component methods

blueprint 13 years ago | |

Interested to read about the better algorithms if you have some good resources.

simonster 13 years ago | | |

Part of the problem with eigenfaces is that they are really sensitive to variations in pose, which, as the author notes, are present in this data set. The article doesn't actually compare the variability among these faces to the variability among a control population of Korean faces, so I don't know to interpret the results anyway, but my guess just by looking at the eigenfaces is that they are capturing pose variation in addition to variation in facial structure.

Check out the results page on the Labeled Faces in the Wild site for some better algorithms that are more robust to changes in pose, along with their performance on a sample data set: http://vis-www.cs.umass.edu/lfw/results.html. Eigenfaces are the worst algorithm tested. Amazon Mechanical Turk is better than all of the algorithms, even if you show only the face itself OR show only the remainder of the photograph that doesn't include the face.

binarysolo 13 years ago | |

Seconding interest on more info in the facial recognition domain and PCA.

hkmurakami 13 years ago |

I do recall a certain cynical Korean friend of mine quipping, "girls who go to the same plastic surgeon look more similar than their own sisters"

sevenatenine 13 years ago |

When you look at them individually they don't look very similar. It's mainly the makeup that highlights the same parts of their faces so when they're flashed all together (eyes in the same spot) it wrongly appears they all look the same.

AYBABTME 13 years ago |

The background of the post aside, I find the main idea interesting and fun. Whatever the validity of the original argument is, I don't think the goal was to actually prove it, but more to play with maths and code to do something cool.

guard-of-terra 13 years ago |

"Recognition" should be replaced by "Morphing" in title.

Currently it makes no sense.

Avshalom 13 years ago |

So... like even in the of-course-it-is-what-are-you-completely-naive photoshopped press photos, they don't really look that similar?

thebadplus 13 years ago |

South korea has a far lower genetic diversity then a lot of european countries, and America for sure. In a totally non-racist way, the "all look same" stereotype may have more truth than it does for a lot of other countries.

seanmcdirmid 13 years ago | |

This is quite easy to test: just take 20 beauty queens from North Korea and see what diversity you get. I believe the genetic diversity in the two countries is about the same, but they have been isolated for around 50+ years, so there is some great science to be had once North Korea opens up (think Iceland).

sliverstorm 13 years ago | | |

"Beauty queens" is a terrible group to sample from. It is not in any way whatsoever random, which is what you would need to study. "Beauty" has many subjective components, which can change rapidly.

seanmcdirmid 13 years ago | | |

But that's not the point. Take 20 girls from each country who are considered by some popular standard to be beautiful, and measure variation. Since I'm assuming plastic surgery is not that widespread in DPRK, the results would be interesting, even if not very useful.

sliverstorm 13 years ago | | |

But you aren't just measuring variation in the population. You're also measuring strictness of beauty standards.

seanmcdirmid 13 years ago | | |

No, we just measure variation according to some algorithm. The beauty is subjective based on standards, and the results would definitely be biased by that subjectivity.

sp332 13 years ago |

"Here we can see that the eigenvalues vanish after 7, suggesting that the rank of the image data is 6."

Does this mean there are only 6 different faces shared among the 20 girls?

eigenvector 13 years ago | |

That's one way of putting it. A rank of six implies that if all of the faces were represented by a matrix (e.g. with 20 rows, one for each face), the dimensionality of the column space would be six. That is, all 20 faces could be represented by linear combinations of six orthogonal faces.

OP implies that six "eigenfaces" faces represent an eigenbasis for the space of all 20 faces (that they are eigenfaces doesn't necessary imply that they form an eigenbasis, not every vector space has sufficient geometric multiplicity to have an eigenbasis).

Wilya 13 years ago | |

It means that the faces can be described using only six uncorrelated parameters. Or to put it another way, there are only six axis of variations. Each face is a combination of the six eigenfaces, but there is still an infinity of different ways to combine these faces, and so an infinity of possible faces.

(Without a comparison to the general population, it's not obvious that 6 is a small number here, so take my "only" 6 cautiously)

flatline 13 years ago | |

I think it only means that each of the 20 faces can be composed from combinations of 6 "basis" faces. I don't know how that compares to the eigenface decomposition of 20 less similar people, I would imagine it is more a byproduct of the process itself than something meaningful?

svantana 13 years ago |

It's all in good spirit, but I can't help to note that it wreeks of the old racist trope "all <insert ethnicity> look the same" http://healthland.time.com/2010/11/24/they-all-look-the-same...

And perhaps also a bit sexist? I dunno, maybe I'm overly sensitive.

mynameishere 13 years ago |

I suspect it helps that these women are absolutely caked in makeup. Plus, averaging out pictures? It gave them an afro. How many Koreans wear afros?

fatjokes 13 years ago | |

it's called an ahjumma perm.

nhebb 13 years ago |

Small correction: I saw this somewhere on the web yesterday, and it turns out that these are Miss Taegu contestants, not Miss Korea.

wjk 13 years ago |

Even with caked makeup and extremely photoshopped pictures they dont even look that similar.