Image Analysis and Verification To Track Pictures

kodiaktau writes "Computer scientist Landon Cox and students at Duke University are working in conjunction with researchers at Microsoft and Technicolor Research to design and develop image analysis tools to identify changes to pictures and videos. The technology, called YouProve, compares original images to create a trust certificate that can be compared against derivative images and produce a heat-map of changes (PDF) between the two. This can be of particular importance when reviewing large amounts of crowd-sourced content to see if image tampering has occurred."

Comments

Watermark detector?

[Anonymous Coward]

Sounds like a watermark detector to me. Probably useful to see differences between image compression as well (artifacts and such)?

Re:Watermark detector?

[Anonymous Coward]

Probably wouldn't be very useful for image compression since almost the entire image would have a slight impression on the heat map (depending on the type of compression, of course.

How is this useful?

[Anonymous Coward]

This sounds more like a DMCA ContentID system rather than a "I can tell this was shopped, I can tell because of this md5 checksum and having seen some shopped images in my time."

Re:How is this useful?

[nescientist]

The alleged utility in verifying crowd-sourced images is a distraction, piggybacking on the occupy wall street thing, from this obvious commercial reality. Microsoft and Technicolor do not give a damn about keeping riot police honest by verifying cell phone pictures. They want to automate the otherwise expensive process of assessing fair use.

I want the firefox/chrome Add-On

[gurps_npc]

Android add-on does not interest me. Make it a Firefox add-on so I can check the pictures I see on facebook/eharmony/j-date/okcupid/match pictures, and you have something interesting.

Re:I want the firefox/chrome Add-On

[MobyDisk]

I bet it is tough to do in Javascript.

Re:I want the firefox/chrome Add-On

[icebraining]

Firefox addons can have binary libraries, which can be called by Javascript.

verifeyed.com

[Anonymous Coward]

There are technologies to test if image has been tampered with even without the trust certificate, e.g. verifeyed.com But I haven't tried it yet to see how good it is.

they're saving a DIFF. big deal.

[Anonymous Coward]

So they are saving a DIFF along with the original image inside its metadata. So what? You need the original and modified image to create the diff. By sending the diff along with the original you're doing nothing other than sending two pictures. I see no use for this at all.

"Trusted" hardware, no thanks

[Anonymous Coward]

The powers that be are awfully worried about people creating content. That could become disorderly.

Solution: lock everything down so people can only consume content within strict confines.

The Orwellian-sounding "fidelity certificate" communicates ownership information.

Welcome to a future where every photo you take can be uniquely attributed to the device used to produce it, registered to you.

When you play in their walled garden you have no privacy. Unattributed images are dangerous, dangerous I tell you. Without a fidelity certificate, how will Facebook know whose advertising profile needs an update?

Re:"Trusted" hardware, no thanks

[RDW]

Welcome to a future where every photo you take can be uniquely attributed to the device used to produce it, registered to you.

Nikon and Canon SLRs have had 'image authentication' for a while now. A supposedly secret private key embedded in the camera's firmware may be used to sign each image file, which can then be validated with a public key (generally using an expensive software package sold mainly to forensics people). Camera serial numbers (probably recorded with your purchase) are also embedded in the files (at least in Nikon raw files), so the original images are intended to be verifiably traceable as well as tamper-evident. However, the authentication feature is optional (off by default) and both Nikon and Canon's systems have now been cracked wide open:

http://www.schneier.com/blog/archives/2011/05/nikon_image_aut.html

You can't stop the serial number being embedded in the file, though (verifiable or not), so be careful who you give original images to...

Re:"Trusted" hardware, no thanks

[Samantha Wright]

A less tinfoil-y way of looking at it might be "Only trusted hardware can possibly produce images with legitimate fidelity certificates. The algorithm for making fidelity certificates is available. Nothing stops me from forging a fidelity certificate after the fact. This is no way to protect an image's fidelity."

Re:"Trusted" hardware, no thanks

[folderol]

Coming next week, an app that will scan your hard drive and remove all identifying metadata from every image file it finds.

Re:"Trusted" hardware, no thanks

[RDW]

Probably a one-liner with exiftool!:

http://www.sno.phy.queensu.ca/~phil/exiftool/exiftool_pod.html#writing_examples

Unfortunately, doing this to Nikon raw files damages them for downstream processing, as the serial number is one of the keys for Nikon's lame encryption scheme that 'protects' white balance data.

Shop Contest

[Thelasko]

Sounds like they are developed a new method to analyze the results of a photoshop contest.

just another failed DRM scheme (its client-trustin

[Anonymous Coward]

This 'new technology' actually requires that every application record what changes it made to the picture. Its ridiculous.

"Cox says the fidelity certificates are produced using emerging tamper-resistant, "trusted" hardware on mobile devices that guarantees they are generated securely and cannot be fabricated. He says that the hardware is a standard feature on PCs and new smartphones but it remains largely unused on both platforms."

Its just another attempt to make DRM which will fail due to someone not actually using the technology and uploading a regular jpeg.

This will kill email hoaxes

[Hsien-Ko]

FOREVER!!!!!!!!!!!!!!!!

Snopes, your days are numbered.

Congratulations, you re-invented fragile w.marks

[hardtofindanick]

Read

Another case of CS guys now knowing about existing EE research.

Difference Layer

[Nom du Keyboard]

And how hard is this compared to keeping a known original and applying it as a difference layer to the suspected altered image? That could look like a heat map.

What am I missing!?

[cwire4]

I'd like to see how well it handles subtler modifications. Rather than blocking out someone's face, what about photoshopping the face like is done in magazines? Like someone else said this doesn't seem much more advanced than a difference layer if you're required to have the original image for comparison. Is the novelty here that it's running on android?

Re:What am I missing!?

[blackest_k]

sounds like another tool to attempt to keep those damn pirates at bay.

I was reading an article at forensic focus about a tool aimed at youtube downloader to identify media that had been format shifted from the original and "pirated" from youtube.

Seems that having a local copy of somethingyou could click and play on youtube is going to be a copyright violation, maybe it is but a pretty trivial one of course if you can get stupid amount of money for each infringing file found on some poor saps hard drive...

http://articles.forensicfocus.com/2011/10/07/youdetect-implementing-the-principles-of-statistical-classifiers-and-cluster-analysis-for-the-purposes-of-classifying-illegally-acquired-multimedia-files-i/

Is it the comet passing by

[Anonymous Coward]

or the occurrence of the words Microsoft and Trust together, I just felt so chilly.

Check out Tineye

[John Sokol]

http://www.tineye.com/ is very cool, you give them a photo and they will search for it, or even photo's that have part of it.

I have some idea of what sorts of algorithms they must be using, but this seems to be a real advancement.

Re:Check out Tineye

[Anonymous Coward]

Yes, Tineye is more useful as it doesn't require the original to be "registered" with a "certificate". The problem with Tineye is that it indexes only a small portion of the web, and they won't even say clearly which sites they crawl. They don't store last-modified dates for the pictures either.

I have a OS-wide tiny tool for you that is better!

[Anonymous Coward]

Have they really just re-invented the "diff" command??

Here is the same thing, but better, using Imagemagick:
composite 1.bmp 2.bmp -compose Difference temp1.bmp
convert temp1.bmp -fill red -tint 200 temp2.bmp
composite 1.bmp temp2.bmp -compose Screen out.bmp

There, done.
And this handy script, when you adapt the coordinates/sizes, and put it on a shortcut, allows you to cheat on any "find the differences" game and get the shortest times ever. (Requires bash, KSnapshot, qdbus, grep, ImageMagick, Gwenview and wmctrl. But it's easily adaptable.)

#!/bin/bash
BASE="/tmp/"
cd $BASE
S="$(date +'%N').bmp"
ksnapshot &
P=$(qdbus | grep ksnapshot)
while [ ! "$P" ]; do
    sleep 1
    P=$(qdbus | grep ksnapshot)
done
qdbus $P /KSnapshot save "$BASE/$S" > /dev/null
qdbus $P /KSnapshot exit &
convert $S -crop 345x452+294+420 1.bmp
convert $S -crop 345x452+641+420 2.bmp
composite 1.bmp 2.bmp -compose Difference temp1.bmp
convert temp1.bmp -fill red -tint 200 temp2.bmp
composite 1.bmp temp2.bmp -compose Screen out.bmp
gwenview $BASE/out.bmp &
while [ ! "$(wmctrl -l | grep -i gwenview)" ]; do
    sleep 1
done
wmctrl -r gwenview -b remove,fullscreen
wmctrl -r gwenview -e 0,1000,400,400,500
wmctrl -r gwenview -b add,above
rm $S 1.bmp 2.bmp temp1.bmp temp2.bmp
while [ "$(wmctrl -l | grep -i gwenview)" ]; do
    sleep 1
done
rm out.bmp

There, a couple of minutes on Slashdot, and we already came up with something better than Duke University and Microsoft is probably weeks.
And this is why I think playing with colorful clickables does not make you a computer user but a appliance user. Automating your work away makes you a computer user. :)

Re:I have a OS-wide tiny tool for you that is bett

[Anonymous Coward]

Here’s a somewhat shortened parallelized and pipelined version of it. It should perform better hard-disk-access-wise and depending on the cost of the sub-shell spawning it's either faster or just as fast. Of course disabling any desktop bling/effects will also help a bit.

#!/bin/bash
BASE="/tmp/"
cd "$BASE"
S="$(date +'%N').bmp"
ksnapshot &
P=$(qdbus | grep ksnapshot)
while [ ! "$P" ]; do sleep 1; P=$(qdbus | grep ksnapshot); done
qdbus $P /KSnapshot save "$BASE/$S" > /dev/null
qdbus $P /KSnapshot exit &
convert $S -crop 345x452+294+420 1.bmp
gwenview <( composite 1.bmp <( composite 1.bmp <( convert $S -crop 345x452+641+420 - ) -compose Difference - | convert - -fill red -tint 200 - ) -compose Screen - ) &
while [ ! "$(wmctrl -l | grep -i gwenview)" ]; do sleep 1; done
wmctrl -r gwenview -b remove,fullscreen
wmctrl -r gwenview -e 0,1000,400,400,500 &
wmctrl -r gwenview -b add,above
rm $S 1.bmp

Re:I have a OS-wide tiny tool for you that is bett

[Anonymous Coward]

Now do it when you only send the edited image to your photo proofing service, adding that the second picture is actually zoomed in slightly.

Re:I have a OS-wide tiny tool for you that is bett

[wzzzzrd]

Bump. Can someone mod this up? That's the stuff I came for to /. in the first place, ages ago. You know, actually interesting ideas instead of this fanboihateboigoldengirlsboi crap that's showing up more and more often.

Planting facade

[domstroi]

Performed planting facade neighbor. He was very pleased. I recommend.

Re:I have a OS-wide tiny tool for you that is bett

[Anonymous Coward]

They don't only send the edited image to the photo proofing service. They also send the diff data. Or the service already has the original or the diff data.
And with existing diff data, it's no different than patching and comparing, as done in every version management system on the planet every day.