Now Pictures On The Internet Can Be Faked

We know it’s shopped, but we can’t tell because of the pixels. PhD student [Kevin Karsch] along with a few other friends will be presenting their methods to render objects into preexisting photos at SIGGRAPH Asia next month.

The paper (PDF…) covers how [Kevin] et al. go about putting impossible objects into photos. The user first defines the geometry of the picture; legs of tables are defined and the table top is extruded from these legs. The lights are then defined by drawing a bounding box and with a little bit of algorithmic trickery, a 3D object is inserted into the scene.

Comparing the results to the original picture is jaw-dropping. For us, photoshopping a bunch of billiard balls on a pool table would take hours, and it would never look quite right. [Kevin]’s work for SIGGRAPH can do the whole scene in minutes and produces results we couldn’t dream of.

There’s no downloadable software yet, but the algorithms are there. Check out the video demo of the techniques and results after the break.


56 thoughts on “Now Pictures On The Internet Can Be Faked

  1. It’ll be interesting legally when this gets to the point that a defense team can claim that any photographic evidence could be fake. It might be the counter to the panopticon society we are headed toward.

    1. IANAL, but there’s a reason I don’t have to deal with red light cameras in my home state.

      The reason red light cameras are illegal in Pennsylvania is because, “You can’t cross examine a camera.” Literally, that’s what the court found (back in the 70s, iirc, and no, I don’t feel like finding the relevant case law or decision). We’re guaranteed the right to cross examine witnesses against us, and it’s impossible to cross examine an autonomous camera.

      I assume that when this tech goes big time, there will be a similar decision in a few more states. Of course that would also getting rid of red light cameras, and those are huge money makers for the county. It’ll be interesting to see how this goes.

      1. On the OT topic of red light cameras, is it really that hard to just stop (or just the concept for some drivers) for a red light and not floor it? I mean… really? reeEAAALLYYYYYY?

      2. calling a camera a witness is idiotic. The photograph is evidence, no need to bring characterization to inanimate objects to swindle some courtroom into making traffic cams illegal. If this is true what you say, “you can’t cross examine a camera, so the evidence can’t be admitted” then this is a massive failure of the justice system.

      3. In the case of a red light camera, the only evidence is the photo. In the hypothetical case of Dan committing homicide, there would be additional evidence upon which to base charges. Due to the ease of tampering with photo and video evidence, many states will not rely solely on photo or video evidence as the basis for a guilty verdict.

        One other point is that red light cameras are not the cash cow many localities assumed they would be. The City of Los Angeles recently turned off their red light cameras because they were losing money on them. There are also studies that indicate that the cameras may not improve safety at red lights.

    2. There needs to be sealed, auto delicately digitaly watermarking camera’s for jouralists use.
      Any altering of the image breaks the watermark in the image, and thus only camera’s with the watermark intack can be trusted.

      While it wouldnt be impossible to generate a new watermark on a digital image, the codes themselves could be secret and one use…garentying a chain of authoticity. So if someone fakes something at least you can tell at what point it was done and thus who to blame.

  2. Surveillance photos of person + fancy math = ‘good enough’ 3D model of person.

    3D model of person + existing photo + math = alibi or damning evidence.

    I’m sure our legal system will do just a super job of sorting it all out though. Judges are great at understanding technology.

    1. Your comment sounds eerily similar to what was said in the video. “Our study shows that each method appears highly realistic to people, and also that our method is preferred to other state-of-the-art techniques.”

  3. I remember when this technique was new and photos of Half Life 2 characters appeared in everyday situations, sat on a comforter in a bedroom, beside a fence in a driveway, sucks that those pics got taken off by the lawyers… some link would appear here

    1. Those HL2 pics were simply photoshopped. These images are not, and that’s the whole point: this technique involves a lone non-artist clicking on a photo to specify the position of lights and surfaces. It evidently takes about ten minutes (not counting the render time) to get the results pictured on the site. They did some trials where they had people look at two scenes, one where the objects are added in using their technique and one where they’re real objects. People picked the fake as the real version 67% of the time.

  4. Unless you’re looking for it they are very convincing. Some of the less texturally complex items still have an ‘uncanny valley’ feel to them, but that could be due to compression losses from the origiinals.

    Beware the weeping angels. Don’t blink, not for a second.

    1. I thought the same thing, but then I saw the side-by-side comparison of the pot, the ball, and the wine glass on the table where the real pic doesn’t look real and realized that my mental image of how reality looks isn’t particularly accurate. The extreme version of this is when people on a forum or whatever call a real photo “obviously faked.” Folks focus on how this tech could let bad guys “prove” things that aren’t true, but I’m honestly more concerned with the possibility of any image being fake allowing us to deny the truth when it’s staring us in the face.

      1. “I’m honestly more concerned with the possibility of any image being fake allowing us to deny the truth when it’s staring us in the face.”
        oh god it will be just like…well…the internet…now

    2. The crystal ball and black bowl on the table are the best. They look absolutely real even when you know they’re faked.

      I think the shiny and flat objects still look fake, the purple pot looks like a solidworks drawing. I think some of the objects were cut out of photos and the edges make them stand out. Some clever anti-aliasing would help fix this.

  5. that could have some far reaching effects.

    lets say you want to have fun with the middle eastern community you could make a picture of osama’s dead body and wreak havoc on the community (the same havoc that was feared being why the president and navy seals are keeping the pictures sealed).

  6. I think that the technological side of the story is excellent. But the problem with this is: what is it really good for? I can think about maybe 5 times more unethical applications than ethical.
    As a professional scientist myself I believe, that these guys should have asked such question before doing this. There are one million things that can be done, but not all ought to be done.

    1. ultimately the technology exists, and to deny it is to favour ignorance to knowledge. are there negative implications? sure. but as einstein said “if knowledge can create a problem it is not through ignorance that we fix it” if he didnt make it someone else would have, its inevitable.
      and its just the start, this is nice and will make it harder to spot fakes and easier to propagate them. it may make pictures worth a thousand lies. so we just have to push through that, find the next technology that makes fakes easier to find, find ways to discriminate between real and fake.
      that’s the wonders of technology and its biggest curse, you HAVE to keep moving forward, there is no retreat, there is no standing still.

      1. I think the argument for mutually assured destruction is essentially if they make it, we have to make it too so they can’t use it on us without us using it on them. Rightly or wrongly this is largely true. If Germany had nukes but the US didn’t and they used them on the US – do you think the war would have gone the way it did?

      2. @steve
        yes i do believe the creation of nuclear weapons was good, and the creation of bioweapons. i disagree with their USE not their creation. science doesnt stand still just because the person who could have invented something chose not to. charles darwin sat on the origin of species for like 30 years, then someone else came along and discovered it just as he did. if one person can make something so can another and its the height of hubris to think anyone can stop that.

        and ultimately nuclear weapons gave us nuclear power, furthered our understanding of atomic theory, and saved many more lives than it took in japan. bioweapons have improved our understanding of biology and especially how to defend against infection diseases. there may come a time soon that deadly viruses can be printed off by people at home for cheap. technology moves at amazing speed and turning a blind eye to the dangers wont make it go away, all we can do is dive in and try our damnedest not to sink.

        its a great time to be alive.

    2. Like most other things at Siggraph, this will probably see a lot of use in the visual effects industry. There have been tons of ways to do this kind of thing in the past, but they have all involved a lot of prep work, finicky camera matching, capture of ambient lighting and partial scene reconstruction by hand.

    3. ” I can think about maybe 5 times more unethical applications than ethical.”

      Your not thinking very hard then.

      This would be crazy usefull for any building visualisation work. Or tv special effects. OR Augmented Reality applications.

      All big industarys.

      On the positive law side crime scene reconstruction too.

  7. While I do think that this technology is great and is probably some of the best faking I’ve seen, it still doesn’t seem “real” to me. Something about the shadows just isn’t right, I guess. It’s impressive, but it still could be improved a lot before everyone starts having trouble telling if a photo is fake. (on a side note, you can tell the real picture in the comparison by looking at the line in the background through the glass, the fakes don’t seem to correctly account for the distortion due to the curvature)

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.