Slashdot is powered by your submissions, so send in your scoop

 


Forgot your password?
Close
typodupeerror
×
AI Technology

Averaging Inanimate Objects Together Produces a Very Human Face 103

Posted by timothy from the what-we-look-for dept.
StartsWithABang writes: It's well known that by aligning and averaging a wide variety of human faces together, an eerie "average" human face can be arrived at. But we see faces in things all the time, from natural scenes like terrain to artificial ones like cars, coffeemakers and combination locks. For the first time, someone averaged together a large number of images of objects appearing to have faces, and the result, strikingly, was an eerily human face. You'd think this might say more about the algorithm than the images themselves, but when noise was used, no human face emerged at all.
This discussion has been archived. No new comments can be posted.

Averaging Inanimate Objects Together Produces a Very Human Face More

Averaging Inanimate Objects Together Produces a Very Human Face

Comments Filter:

  • Surprise! (Score:5, Insightful)

    by Anonymous Coward on Sunday November 15, 2015 @08:10PM (#50936855)

    So, average a bunch of things that only have their resemblance to a human face in common, and you end up with a human face? I didn't see that coming.

    • I think people are just wired to seek out humanity in things.

      It's not supernatural, just how the brain is wired thus far.

      • Re: Surprise! (Score:5, Insightful)

        by PopeRatzo ( 965947 ) on Sunday November 15, 2015 @11:29PM (#50937535) Journal

        I think people are just wired to seek out humanity in things.

        But strangely, not in each other.

        • Re: (Score:1)

          by Anonymous Coward
          Even if you are insightful to no one else, you are insightful to me

        • I'm not sure why "strangely"

          One reason we're so good at seeing faces everywhere is probably to better avoid predators looking directly at us. (Whether they're other animals or human). Now, we do see humanity in others, it's surprisingly difficult to train a human to kill others, but it's a much weaker effect. And from an evolutionary standpoint it probably must be. If we didn't have it, the genes for sociopathy etc. could spread uncontrolled, as there'd be no defense against them. As it is, we're keeping st

      • Re: (Score:2)

        by sudon't ( 580652 )

        Right, I don't get what his point is. We have a whole section of our brains dedicated to recognizing faces, so naturally, we experience pareidolia all the time.

        "Imagine that: a human face, emerging from the averaging of inanimate objects like combination locks, metal finishes and coffeemakers. And yet, there it is, a face that’s recognizably human. All it takes is a pretty remarkable combination of psychology, design and technology, and shadows of ourselves begin to emerge."

        It's got nothing to do with "psychology, design and technology". It has to do with how our brains are hard-wired.

    • Essentially, this method should show what kind of traits look like faces to us rather than what real human faces look like. It's exploring properties of the psychovisual system of humans, not properties of face detection algorithms or statistical human faces.
    • While true, I thought the interesting part was WHAT the face looked like.

  • No kidding (Score:5, Insightful)

    by shellbeach ( 610559 ) on Sunday November 15, 2015 @08:12PM (#50936865)

    You'd think this might say more about the algorithm than the images themselves, but when noise was used, no human face emerged at all.

    Wait, so, when images that looked more like faces were used, the average looked more like a human face? Just crazy.

    It's cute, but I'm not sure it's particularly profound.

    • Comment removed based on user account deletion

      • Re: (Score:2)

        by Sique ( 173459 )
        Combinging 50 sounds which sound like a bird might not sound very birdlike. You might end up with some kind of white noise.

        • Not random: Faces Aligned and Similarly Sized (Score:5, Informative)

          by Roger W Moore ( 538166 ) on Sunday November 15, 2015 @10:03PM (#50937237) Journal

          Combinging 50 sounds which sound like a bird might not sound very birdlike. You might end up with some kind of white noise.

          Probably true but I bet if you took images of human faces which were not already aligned and not all zoomed to a similar size then that too would generate noise. The only reason the averaging works is because people naturally take photos with the face the right way up and zoomed to a similar size. I bet if you were allowed to do the same alignment and scaling for bird song you could average the now aligned audio to get something like birdsong.

          This is why this result is so obvious and not at all what it says. These are not random face-like images but ones with the same alignment and comparable zoom factor. If I did the same for any shape I would get the same result: the details of the shape would blur but the basic shape would remain the same because they are all aligned and have similar sizes. Someone should nominate this for an ignobel prize.

          • Re: (Score:1)

            by rgmoore ( 133276 )

            I bet if you were allowed to do the same alignment and scaling for bird song you could average the now aligned audio to get something like birdsong.

            I am not nearly so confident. Maybe if you averaged the song of many birds of the same species you could get some kind of recognizable song out. But what's going to happen when you average the song of a chickadee, a robin, a meadowlark, and a crow? There's simply no way of aligning them so they produce a coherent combination; they're just too different.

            • Seems possible with Auto-Tune...

            • Maybe if you averaged the song of many birds of the same species you could get some kind of recognizable song out.

              Exactly - but that is what they are doing for faces. They are not averaging human, ape, bird, spider, insect etc. faces but the faces of a single species: humans. So by analogy it is perfectly reasonable to specify the same species of bird and then adjust the frequencies to match (since size variation will affect the frequency) and then add an appropriate delay so they all start the same part of the tune at the same time. This is exactly what the OP did with the images and I would agree that with the same

          • In fact, in TFA it sounds like the photos were not necessarily aligned by the photographers, but the person who did the averaging also aligned them beforehand. So yeah, nothing too astounding here. However, it does perhaps give us a blurry but explicit idea of a basic imprint of "faceness" that we implicitly look for when determining whether an object seems to us to have a face. This could be interesting. Maybe. Not really.

      • Well, it could be interesting. Look at it this way, maybe they've tested thousands of algorithms before where the results were disappointing blobs and now they managed an averaging algorithm that does what we intuitively want it to do.

    • It's cute, but I'm not sure it's particularly profound.

      Well, this is obviously leading up to the discovery that you can average pictures of toast, windows, dog butts and what not to get a picture of Jesus.

      • Re: (Score:2, Funny)

        by Anonymous Coward

        But can you can average pictures of Jesus to get toast? Perhaps the feeding of the five thousand is a distorted recollection of an image processing course. My theologian friend always tells me a lot is lost in translation.

    • The fact that you get /a/ face isn't profound, but the resulting image is interesting. It gives a good picture of the things that human vision uses to locate faces: obviously the eyes and mouth are most prominent; there's moderate contrast for the cheekbones and nose; the oval shape is only vague; the neck, ears, eyebrows, and hairline are almost entirely missing.

      I expect those are already well known to vision specialists, but to me, it's an interesting analysis of the exact details which make an inanimate

      • Re: (Score:2, Interesting)

        by Anonymous Coward

        They used face detection algorithms to align the pictures.
        The only thing we know is that the algorithm aligned the eyes, mouth, cheekbones and nose. The other features will appear at varying positions.
        Had the face recognition algorithm used ears and eyebrows to align the pictures those would have been the most prominent.

        You can generalize this method to anything. Take a random image and rotate, scale and align it so that it matches your reference image the best.
        With a macroscopic amount of input images, eve

        • Re: (Score:2)

          by GuB-42 ( 2483988 )

          The interesting part in this study is that they tried with random data and the result didn't look like a face at all, even though the filtered sample size was greater (47 vs 16 detected faces).
          If means that there is more to it than just the algorithmic criteria.

    • Re: (Score:1)

      by Anonymous Coward

      Next you'll tell me when you average random numbers between 1 and 100 you get something close to 50. Now that's just freaky!

  • Go away, startswithabang (Score:5, Insightful)

    by Anonymous Coward on Sunday November 15, 2015 @08:12PM (#50936867)

    Your blogspam links to Forbes is offensive.

    Stop linking a website that is designed to broke hyperlinks and to force through an ad page.

    • More like "StartsWithACrap" because the site is nothing but bad science and shite sensationalism

    • Not to mention the page itself which is laden with ads. Plus the rubbish article. I never found Forbes a great magazine to begin with, something for a PHB to read on the airplane to appear intellectual, but it's really gone to the dogs lately judging from the last few articles linked here.
    • Well said. All I see is a grey page. Does anybody have a direct link to the actual image obtained?

  • So averaging things which look like a face... (Score:5, Funny)

    by UpnAtom ( 551727 ) on Sunday November 15, 2015 @08:32PM (#50936933)

    Did the rest of the world suddenly get dumb?

    Here averaging is basically:
    What's common amongst all/most of these images that look like a face?

    Yes, it's something that looks like a face. Shocked, I tell you.

    • Re: (Score:2)

      by MacTO ( 1161105 )

      Maybe. Maybe not.

      I suspect what they're saying is: take a bunch of inanimate objects that are identified by an algorithm as having the properties of a human face then average them together. What you end up with is something that looks more human than any given image.

      Then again, that probably isn't surprising. You would expect the algorithm to identify things as human faces if the computed values are within some range, with the range being roughly centred on the average. The natural consequence being tha

      • Re: (Score:2)

        by AK Marc ( 707885 )
        The algorithm locks on to a face, and rotates, sizes and such to make the faces line up. The thing is that the faces averaged look very human, not like a cat, or smiley face. Is that the algorithm, or the way we build faces into our objects?

    • Did the rest of the world suddenly get dumb?

      That's what happens when you're in hypersleep for 57 years.

  • there's this facebook thing, too. when you gaze long into an average......

  • forbs (Score:2, Insightful)

    by JohnVanVliet ( 945577 )

    man that is one F'ed up site that i can not get past the ADVERTISING 3 second countdown

      that is royally F'ED UP

    • Whenever I see a Forbes link redirect to forbes.com/forbes/welcome, I click back (page doesn't even need to fully load) and click the link again. The second time it goes straight to the target. Works with both Firefox and Chrome w/ ABP.

    • Re: (Score:2)

      by lkcl ( 517947 )

      man that is one F'ed up site that i can not get past the ADVERTISING 3 second countdown

      that is royally F'ED UP

      you get an advertising countdown? all i get is a redirect to a URL with the word "welcome"... that's then completely blank. this isn't the first time it's happened, either. y'know what? i'm going to resurrect a tag that i haven't seen used in a looong time - i'm going to mark this artlcle as "slashdotted". yes, in 2015, that's an indictment of a site as big as forbes. in 1997 it used to be fine (and funny) that the 15 minutes of fame from a slashdot focus would overwhelm a web site link.

  • Pick 15 things that look like a face
    average them together
    result looks like a face

    What is happening to Slashdot?

    • Re: (Score:2)

      by AK Marc ( 707885 )
      A car grill that looks like a face is averaged with a toaster and others and it comes out looking very human. That's the issue.

      • I'm no expert in human cognition or neural nets, but it seems to me that if, on average, something looks like a human face to the average human, then it must share regular correspondences to a human face. Take a few of these that vary in one way or another but all share such average correspondences and average them together, one would expect to get a human face.

        I don't know, the obviousness factor is just still there for me. Maybe the profundity escapes me.

    • Well, obviously: If you're averaging many real numbers between 41 and 43, you'll get a number close to 42. If you're averaging many "objects that look like human faces" you'll get something that looks like a human face. What did they expect ? White Noise ? A porcupine ? 7 ?

    • Well, if you average out all the articles on slashdot recently, you'd likely get goatse. Or maybe a dice job listing.

  • Forbes (Score:3)

    by kamakazi ( 74641 ) on Sunday November 15, 2015 @10:50PM (#50937401)

    I really hate to contribute to the hate noise the haters bring, but I really hate to visit websites that hate to let me see the site without allowing scripting I hate from dozens of hated sources.

    Could we get some kind of automated indicator when a link points at a site that just won't load with NoScript?

    I don't think I am a tinfoil hat paranoid, I just don't like to have to allow 17 different sites to run scripts in my browser just to read an article. After reading a few comments it looks like I didn't miss much this time.

    • Re: (Score:3)

      by cas2000 ( 148703 )

      Slashdot should go beyond just warning about such crap sites, they should ban linking to them.

      A web site that insists on being able to run un-vetted, untrusted code on your computer just to display some text and pics in an article does not deserve to be trusted.

  • Just great (Score:5, Funny)

    by Anonymous Coward on Sunday November 15, 2015 @11:11PM (#50937469)
    Next thing you'll be telling me that when you average 50 photos of assholes you'll come up with something that looks remarkably like [pick your favourite politician].

    • Re: (Score:2)

      by cas2000 ( 148703 )

      or average the pics of 50 top businessmen and you'll get something that looks remarkably like an arsehole.

  • I've just averaged all the human faces together and got an early '60s Jaguar E-Type:

    http://image.motortrend.ca/f/1... [motortrend.ca]

  • It got a real scam site ... "continue to article in 3 2 1 ...". Are we talking about a warez site or a news site? Would not click again.

  • 2spooky4me [forbes.com] Usually average faces are considered attractive, not eerie. And of course they're just called eerie in the Slashdot summary, not the article.

  • They consciously trained their algorithms to do what babies do: pan, and zoom/frame-in-software until facelike features emerge. But what is more interesting, the research relies on YOU to take the last, great leap.

    They present the results --- the one from merged faces and the one from objects --- as a 'face'. But is it really a face? Ask yourself, if you were walking outside under perfect lighting conditions and someone with the precise blurred faces shown as their result approaches you. What would your rea

Slashdot Top Deals

An Ada exception is when a routine gets in trouble and says 'Beam me up, Scotty'.

Close