Hacker News new | comments | show | ask | jobs | submit login
Korean Beauty Contestants – Face Morphing (jbhuang0604.blogspot.com)
173 points by thebadplus 1606 days ago | hide | past | web | 55 comments | favorite

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


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

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

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.

That's a nice point about points.

surgery first, then makeup, then lighting, then photoshop

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

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

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

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.

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.

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

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 :(

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.

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.

Look here:


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.

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

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

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

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

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.

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

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"

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.

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.

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

Currently it makes no sense.

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

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.

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).

"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.

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.

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

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.

"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?

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).

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)

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?

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.

It's not a racial thing. See these translated Korean netizen comments: http://netizenbuzz.blogspot.com/2013/04/foreign-media-outlet...

> 1. [+238, -4] They look the same even in our eyes, I can't imagine in the eyes of foreigners... -,.-

> 2. [+234, -9] ㅋㅋㅋㅋ I thought all the faces were ctrl + C and ctrl + V with just the shirts changed. They should be ashamed of getting plastic surgery, how could they even think of entering a beauty contest? Seems like plastic surgeons can surgically remove shame now as well..

> 4. [+31, -1] Just turn Miss Korea into an exhibition contest for plastic surgeons

> 5. [+18, -2] I'm so embarrassed... They should make sure in the next Miss Korea that they only allow natural girls

> 10. [+17, -0] That's why they're called 'docturnal twins'. Twins made by plastic surgery doctors.

Internet commenters slagging off beauty contestants as identical barbies? News at eleven.

No, it's that people are getting het up about this particular instance that feels a bit... xenophobic at least. Especially because of the apparent credulity that a bunch of press photos was received with.

Jeez, can't a guy read an article about principal component analysis on a published sample of photos without being painted as some kind of racist xenophobe?

It's hardly the first time those looking for something to take issue with have found it in digital imaging: https://en.wikipedia.org/wiki/Lenna

The problem is that "they all look the same" is actually a factual observation about how our brains recognize faces. So, if you label it as a "racist trope" you end up implying that racism is part of human nature... which I'm sure is not what you want.

This wouldn't be the first time "basic human nature" flys at odds with equality movements.

In this case, it would be an inequality movement

This experiment doesn't seem to be positing at all Koreans look the same but rather that all Korean beauty contestants look he same. The implication being, not that Koreans look similar to each other, but that they have similar standards of beauty. Apparently, most of these women have had plastic surgery so they all seem to have the same ideal when it comes to their face. I'd like to see a Brazilian (another country with a large plastic surgery industry) equivalent to see if it's the same.

Rather, in this sample of 20 photos the contestants look similar, but "contestant 1, 2 and 6 are more distinct than the rest of the contestants".

I thought he did pretty well to not draw any conclusions in the article.

When it's actually got a foundation in math, it stops being a racist trope and becomes a valid scientific inquiry.

Sexist? Overly sensitive would be one word for it.

The analysis uses math, but that doesn't make it scientific. In particular, there's no comparison to the similarity of a sample of faces of normal Koreans, westerners, or beauty contestants.

Unless the "foundation in math" is really confirmation bias, which math as used by humans is still vulnerable to.

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

Well we are talking about Korea. You would probably have a heart attack if you watched even 60 seconds of Kpop.

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?

it's called an ahjumma perm.

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

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

Applications are open for YC Winter 2018

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact