Slashdot Mirror


DARPA Program Targets Image Doctoring (networkworld.com)

coondoggie writes: It isn't hard for just about anyone to change or alter an image these days — and that can be a problem. It's an issue researchers at the Defense Advanced Research Projects Agency want to put to rest with a new program called Media Forensics, or MediFor, which looks to build an algorithmic-based platform that can detect image manipulation. "The forensic tools used today lack robustness and scalability and address only some aspects of media authentication; an end-to-end platform to perform a complete and automated forensic analysis does not exist. Although there are a few applications for image manipulation detection in the commercial sector, they are typically limited to a yes/no decision about the source being an "original" asset, obtained directly from an imaging device. As a result, media authentication is typically performed manually using a variety of ad hoc methods that are often more art than science, and forensics analysts rely heavily on their own background and experience," DARPA states.

41 comments

  1. Frosty by Hognoxious · · Score: 4, Funny

    Just check if some of the pixels are wrong. It helps if you've seen a few shops.

    --
    Confucius say, "Find worm in apple - bad. Find half a worm - worse."
    1. Re:Frosty by invid · · Score: 4, Insightful

      If you can create an algorithm that can detect pixels that have been modified in a picture, you can create an algorithm that can modify the pixels to hide the fact that they've been modified.

      --
      The Moore-Murphy Law: The number of things that will go wrong will double every 2 years.
    2. Re:Frosty by jbmartin6 · · Score: 1

      Didn't Captain Kirk use this idea to cause an evil computer to explode?

      --
      This posting is provided 'AS IS' without warranty of any kind, implied or otherwise.
    3. Re:Frosty by dcw3 · · Score: 1

      If you can create an algorithm that can detect pixels that have been modified in a picture, you can create an algorithm that can modify the pixels to hide the fact that they've been modified.

      So, are you able to reverse https://en.wikipedia.org/wiki/... I'm sure it's possible, like MD5, but not likely for all but groups like NSA.

      --
      Just another day in Paradise
    4. Re: Frosty by Anonymous Coward · · Score: 1

      Is society trying to eliminate anyone have specialized training or knowledge, or is that just a side-effect of its downward spiral?

    5. Re:Frosty by Anonymous Coward · · Score: 0

      Are you thinking of when Captain Picard got captured by the Borg and implanted with a visual device like their's; then the Enterprise folks figured out an impossible shape to feed the visual processors which caused them all to "hang" and enabling a rescue? Kirk ~= Picard, and Evil Computer ~= Borg.

  2. Other competitors ? by Anonymous Coward · · Score: 0

    I know of tungstene which is used by some press association.
    Do you know other ? Possibly FOSS

  3. Signing cameras by Chaset · · Score: 1

    I don't know about identifying things after the fact, but an idea I've tossed around for a few years now is a forensic digital camera. Basically, the hardware will sign/watermark all the photos it takes with some sort of digital signature unique to the camera. The private key would be buried in silicon in such a way as to destroy it if attempts are made to discover it. I'm not a security/encryption expert by any means, so I don't know how feasible this is (or does it already exist?) but sounds plausible to me.

    --
    -- "This world is a comedy to those who think, a tragedy to those who feel."
    1. Re:Signing cameras by Anonymous Coward · · Score: 1

      That functionality already exists and has done for some time.

      However the implementations so far haven't stood up to attack - Canon Nikon

      There's no reason that should be the case though. It should be something that can be done securely, only falsifiable if you can either crack the key or find a hash collision (which'd likely mean making enough changes to the image to make it obvious that it's been modified).

    2. Re:Signing cameras by Anonymous Coward · · Score: 0

      You would also need a database of trusted digital cameras for this to work. And you only need a single compromised camera to skip all the benefits this gives.

      It would also not guard against e.g. taking a picture of a picture. You could include a lot of sensor data that could be used to detect such cases, but it would make the system significantly more complex than the easily verifiable signing/watermarking process.

    3. Re:Signing cameras by dsmatthews9379 · · Score: 1

      You can't mandate what cameras people use, but you can have people submit images to a digital signing service to prove the image existed in that form at that point in time, this would prevent anyone from using it as source component of a forgery.

      The problem is that I can otherwise create a virtual camera in software that signs the forgery after it is created. I can even use wavelets to extract the grain/noise/high frequency artefact layers from an image before I modify it, then return them to obscure "identity difference" in parts of the final image. You can even calculate the lighting model from an existing scene to ensure the added components match perfectly.

      If you are trying to detect more than some clowns Photoshop fake it is actually a very difficult problem.

    4. Re:Signing cameras by Actually,+I+do+RTFA · · Score: 1

      I can otherwise create a virtual camera in software that signs the forgery after it is created. I can even use wavelets to extract the grain/noise/high frequency artefact layers from an image before I modify it, then return them to obscure "identity difference" in parts of the final image. You can even calculate the lighting model from an existing scene to ensure the added components match perfectly.

      Is that an "I can" or a "One can"?

      --
      Your ad here. Ask me how!
  4. Algorithm: by Edis+Krad · · Score: 1

    if ( CanTell(somePixels) && EnoughShopsSeen( time )) {
    printf("THIS LOOKS SHOPPED");
    }

  5. Shopped, not doctored by Anonymous Coward · · Score: 0

    Are you from the past?

    1. Re:Shopped, not doctored by ensignyu · · Score: 1

      Hey, I didn't spend four years in image doctoring school to be called a shopper!

  6. Silly by Anonymous Coward · · Score: 0

    Okay, let's say it works out somehow, with not too many false positives, but most likely too many to hold up in court anyway. It'll work really nicely, until the other parties find out how the algorithm works, and then slightly change the way they manipulate images in some way.

  7. Re:Bernie Sanders 2016 by Anonymous Coward · · Score: 0

    Everyone with two brain cells is already rooting for Trump.

  8. I hate the (tech) world I live in... by NoNonAlphaCharsHere · · Score: 1

    The one where "endtoend" appears as one word, but "ad hoc" appears as two - in the same quote.

  9. By definition....if you can see it, you can fix it by Maxo-Texas · · Score: 1

    nt.

    --
    She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
  10. Re:Bernie Sanders 2016 by Anonymous Coward · · Score: 0

    That'd be the end of civilization for mankind and the beginning of the next dark ages.

    Then again, it's not going to happen anyway because Trump could never win an election and the Republicans know that.

  11. Authentication? by lbmouse · · Score: 1

    I'm still waiting for the "Enhance Button".

  12. Re:Bernie Sanders 2016 by tepples · · Score: 2

    Everyone with two brain cells is already rooting for Trump.

    Fortunately, I have a lot more than two brain cells, so I can see how Trump will repeat history in a bad way.

  13. Adobe doesn't like use of its brand as a verb by tepples · · Score: 1

    Or has the OP been contacted by counsel for Adobe Systems?

  14. Re:By definition....if you can see it, you can fix by tepples · · Score: 1

    The last time this was trotted out, it was something you can't see: steganography.

  15. I can tell cheap photoshop color adjustments by 140Mandak262Jamuna · · Score: 1

    I have noticed that if you use Photoshop to do enhancement of contrast/brightness etc the color histogram will have lots of blanks and will look like the skyline of Manhattan not the ridge line of Rocky mountains. Especially if I use something called "gamma correction" (hope I remembered this term right). Is that a characteristic of all image processing? or just the implementation "feature" of Photoshop.

    --
    sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
    1. Re:I can tell cheap photoshop color adjustments by Anonymous Coward · · Score: 0

      Histogram equalization works by taking a bookshelf with only 6 books on it and spreading those 6 books out so they fill the entire bookshelf. You could make it less obvious that it has happened by taking the equalized image and randomly redistributing the colors to fill in the empty bins...

    2. Re:I can tell cheap photoshop color adjustments by spitzak · · Score: 1

      Error diffusion will help (calculate the floating point value and then store the two nearest integers with a probability based on the value) but not if the correction is extreme. Blurting the image to a floating point value will fix it but, of course, introduces blur.

  16. Hard without Image Understanding by Anonymous Coward · · Score: 0

    The actual BAA is at: https://www.fbo.gov/spg/ODA/DARPA/CMO/DARPA-BAA-15-58/listing.html
    Slides from the Proposer's Day: http://www.darpa.mil/attachments/MediForProposersDaySlides.pdf

    Program wants to identify 'doctoring' in different ways, but i'm not sure they are all feasible. the first one, digital integrity makes the most sense, as you can find changes in compression, copied areas (using histograms or other techniques), edges. Here's an example: http://www.ws.binghamton.edu/fridrich/Research/copymove.pdf

    The second way is physical integrity, looking at shadow angles and lighting. I am not sure how you can do that without having an understanding of the image. Maybe you can automatically determine different shadow angles and make sure they are all consistent? Maybe you can determine if something is moving based on motion blur and then determine something else? Sounds really hard.

    The third one sounds impossible: the 'semantic integrity'. You have to have a human-level understanding of the contents of the picture to determine if something is out of place. If you see an F-35 aircraft in a WW2 photo, for example, you (a person) can say that it has been doctored, but for the computer to do that, you'd have to recognize when the picture was taken based on content of the image and then recognize the F-35 and know its historical range doesn't cover the photograph. Doing location is even harder: a Panzer II in a Iwo Jima picture Hm...that doesn't sound right, but you'd have to know who was fighting where and when.

  17. Novel Explores this Idea by invid · · Score: 3, Interesting

    Warning: Shameless Self-Promotion

    I've written a science fiction novel, The NPC that deals with the ramifications of this sort of thing. The solution in the novel is extreme: all recording devices are required to stream their data to a trusted 3rd party (in this case, a corporation called VuDyne) in real time with an encrypted certificate. Otherwise the digital data is not trusted to represent reality. As you can imagine, this gives VuDyne a great deal of power.

    --
    The Moore-Murphy Law: The number of things that will go wrong will double every 2 years.
    1. Re:Novel Explores this Idea by Anonymous Coward · · Score: 0

      Why not just have each device add an encrypted checksum using public key encryption. That way you don't have to send all the data to one source, the verification can be done by anyone and as long as the encryption isn't broken any tampering is detectable.

      It also doesn't give a ton of power to one verification entity.

    2. Re:Novel Explores this Idea by The-Ixian · · Score: 1

      Well that doesn't make for a very fun "evil corporation" story, now does it?

      --
      My eyes reflect the stars and a smile lights up my face.
    3. Re:Novel Explores this Idea by invid · · Score: 1

      Why not just have each device add an encrypted checksum using public key encryption. That way you don't have to send all the data to one source, the verification can be done by anyone and as long as the encryption isn't broken any tampering is detectable.

      It also doesn't give a ton of power to one verification entity.

      The company also provides the service of enforcing copyright laws.

      --
      The Moore-Murphy Law: The number of things that will go wrong will double every 2 years.
  18. Re:By definition....if you can see it, you can fix by Anonymous Coward · · Score: 0

    The last time that was trotted out, it was called DRM

  19. It can't work in the long run by Baron_Yam · · Score: 1

    If you can use an algorithm to detect tampering, you can use that same algorithm to alter your image so the algorithm no longer detects the manipulation.

    1. Re:It can't work in the long run by dcw3 · · Score: 1

      If you can use an algorithm to detect tampering, you can use that same algorithm to alter your image so the algorithm no longer detects the manipulation.

      See my earlier response regarding https://en.wikipedia.org/wiki/... and tell me how you'd get around it.

      --
      Just another day in Paradise
    2. Re:It can't work in the long run by Baron_Yam · · Score: 1

      Your previous comment is the same as this one. SHA-2 has nothing to do with it - you could simply encrypt your altered photo. If you wanted to do it the (physically) hard way (assuming you're talking about a camera with internal encryption hardware), you could wire something up to the CCD inputs so the camera doesn't realize it's being fed a pre-made image instead of a view of the real world.

    3. Re:It can't work in the long run by dcw3 · · Score: 1

      Okay, I think we're misunderstanding each other.

      My point...If I take an original image, and apply an encryption to it (this doesn't necessarily involve doing so with the camera), and that's my original image which I post for the world to see. Someone else can't necessarily come along and mess with that.

      Yours...Sure, if you're the originator of some photo, and you've doctored it, I agree with you.

      --
      Just another day in Paradise
  20. Like those photoshopped Al-Qaeda pics?? by Anonymous Coward · · Score: 0

    You know, when the hacker conference inspected the images released and found that the pictures had the Al-qaeda symbols added to them at the same time with the same program as the Texas PR company that supplied them? ;)

    I guess if you had a way of testing the photos before releasing them to the public you could avoid the embarrassment of the public finding out that not only were the pictures faked, but about the company that made them, and the "Muslim" actor who turns out to be the son of a Jewish lobbyist...

  21. What I want to know is... by Anonymous Coward · · Score: 0

    Will the product of this have detected that the satellite photo evidence from Iraq for WMD usage was a lie? Oh, wait, that was twisting the interpretation of the photos. Maybe we need a tool for that instead, before getting into the next war.

  22. What if... by feufeu · · Score: 2

    ...the doubts on the reality of any kind of imagery cannot be overcome and we need to abandon the idea that images (moving or not) can be trusted as evidence ? Would the world stop spinning ? I highly doubt this.

    Perhaps there can be an exception in cases where the entire chain of taking and handling an image can be verified in one way or another ?

    Unalterable checksum produced by the camera perhaps ? I know that we can already do this with GPS flight logs (track/altitude) coming from certified flight recorders. (See http://www.fai.org/gnss-record... ) It would certainly require certification of the camera used for taking such an image.