Adobe Tackles Photo Forgeries

Several readers wrote in with a Wired story about the work Adobe is doing to detect photo forgery. They are working with Canon and Reuters (which suffered massive bad publicity last year over a doctored war photo) and a professor from Dartmouth. (Here is Reuters's policy on photo editing.) Adobe plans to produce a suite of photo-authentication tools based on the work of Hany Farid (PDF) for release in 2008.

Comments

Linky

[Anonymous Coward]

PDF is boring. HTML is awesome. Here's the work of Hany Faid in HTML, courtesy of Google.

Staged Photographs

[Detritus]

Besides image manipulation, there is also the problem of staged photographs, as seen in some of the photographs from the recent war in Lebanon. This can't be solved with technology.

Re:Staged Photographs

[TheCodeFoundry]

Which photographs?

I would start with this:
http://rayrobison.typepad.com/ray_robison/2006/08/ al_ap_at_it_aga.html/
and
http://littlegreenfootballs.com/weblog/?entry=2211 6_Still_More_Photo_Staging_Identified&only/

Then you can read about the Red Cross Ambulance Incident here:
http://www.zombietime.com/fraud/ambulance/

Additional staged incidents here:
http://littlegreenfootballs.com/weblog/?entry=2212 3_Green_Helmet_Admits_Staging_Photos&only/

Re:Staged Photographs

[c_forq]

Back in the US invasion of Cuba good old Teddy Roosevelt had bodies moved from one battle front to the one his Rough Riders were on for photographic purposes. There are also incidents of a famous civil war photographer having multiple pictures of the same corpse in different poses in multiple locations. This isn't anything new and it will probably never go away as long as photography is an effective medium of communication.

Re:Staged Photographs

[Anonymous Coward]

RE: LGF -- they do screw up occaisionally but they also admit it when they have screwed up and publish the appropriate retraction. In terms of accuracy, it's a decent source for news, but you can't cherry pick articles from it no more then grabbing a single newspaper off the shelf of an archive would have you miss the correction to the story published next week.

On the other hand if you don't like the fact that it's a politically conservative site about the current state of the world and documenting people set on fire in the name of Islam, that's fine too. But don't say it's inaccurate, merely that you disagree.

As far as the photos are concerned, I don't think it's open to debate. The facts are that the photos were published and presented to the reader as accurate representations when they really depicted staged and altered scenes.

Uhhh, perhaps some non-biased humans are needed to

[SengirV]

How is Adobe going to find other faked war photos like these?

http://zombietime.com/reuters_photo_fraud/r1891896 384.jpg
http://zombietime.com/reuters_photo_fraud/r3577351 291.jpg

or the woman who shows up to cry over every and all bombed buildings in Reuters' world

http://zombietime.com/reuters_photo_fraud/beirutwo man2.jpg

Source - http://zombietime.com/reuters_photo_fraud/

Re:You know what would be cool...

[Joe Decker]

Canon's DSLRs do checksum the data, there's a verification tool as well. Of course that only works with the original uncropped data, but it does give you a fairly firm reference to which you can compare any derivative versions.

Re:Matching images to cameras

[Joe Decker]

Too late, It's already done. The Exif information from the cameras I use already includes the camera serial number. (Not that I'm disagreeing with your point.)

Re:Matching images to cameras

[bustersnyvel]

Too late, It's already done. The Exif information from the cameras I use already includes the camera serial number. (Not that I'm disagreeing with your point.)

Of course, EXIF contains a lot of information about your camera. However, the data is digital, and can thus be edited. You are free to remove any identifying data from the EXIF headers before you publish your images.

Re:Matching images to cameras

[Tophe]

Only jpg files support Exif so saving the picture as a png (or other format) will eliminate the Exif data very quickly and easily.

Re:Matching images to cameras

[Hijacked Public]

All RAW formats I know of do as well, as do Adobe's various ones.

Regardless, EXIF is easily edited and tells us little to nothing about the original image's authenticity.

Re:Doctoring? Yes.

[phlinn]

Not convincing. You glossed over the upper left section of smoke, among other things. There was nothing there before hand, it was added, and the same pattern on the left side is obviously repeated. There are obvious buildings added in the editing photo that aren't there in the original. You point to a building at 2c and 2d in your file which is cloned to 3a and 3b. However, the one at 3a and 3b can be seen in the original, but was moved down to the lower section. More importantly, it's not at quite the same relative postion within your gridlines. Shifting down a bit, and over half as much is very plausible, and since it's not actually regular, your argument is completely unconvincing.

The whole lower half of the original appears to have been copied, sharpened, copied back in lower and to the left, and the smoke added in a vain attempt to cover it up, then cropped to hide the lower right corner which didn't have anything in it. The contrast was increased as well, which definitely makes for a more jarring image.

Re:Doctoring? Yes.

[BlueStraggler]

Methinks thou dost protest too much. This image is faked to a degree that only an incompetent human being could fake. The technical minutiae of the particular method of fakery is beside the point - to my eyes it looks more like a pattern fill than a clone stamp (due to the regular repetition you note), but we could argue about that all day. The dead give-away that unscrupulous human beings are to blame are to be found at the edges of the doctored areas. No general-purpose algorithm is going to expand the cloud of smoke preferentially in one direction, and then suddenly terminate the billowing edge of the smoke cloud against a clear sky, because general-purpose image enhancement algorithms do not model the behaviour of billowing smoke clouds. No general-purpose algorithm is going to cut out whole buildings and transplant them perfectly to other parts of the neighbourhood, because general-purpose algorithms do not recognize where buildings start and end against a backdrop of other buildings. It takes a highly advanced image processing tool (namely a human being) to select meaningful subsections of an image (a particular building, a particlar part of a cloud) and reproduce it somewhere else in the image that makes sense to an intelligent viewer.

In other words, if a generic photoshop filter were to move buildings around the city, and enhance billowing smoke clouds in such a way as to enhance just the cloud without randomly chopping up other parts of the image, as was done in this image, then we could conclusively state that we have achieved artificial intelligence in commercial software. The fact that the result was lame is moot, because the necessary filters to clean up/smudge the lameness are dead easy, compared to the filters that made the initial image edits.

But the fact that a 10-year-old (or someone with equivalent aesthetics) could have made those photo edits in 10 minutes seems a somewhat more plausible explanation than the notion that we have HAL 9000 embedded in Photoshop.

Re:Uhhh, perhaps some non-biased humans are needed

[Anonymous Coward]

That was the point. The photos weren't from Zombie. They were from photographers/stringers over in the Middle East. In this case, I believe it was from Lebanon, during the recent war involving Hezbollah and Israel. Zombie inspected several of the more poignant images that came from the war and were posted on Reuters. She's (I believe it's a her) claiming shenannigans.

In case you don't understand (rather than making a rhetorical comment), the toys ARE suspiciously clean. Which, if it were in a builing that had been bombed to rubble... makes little sense. Along with the other toys and childrens' articles. That's one of Zombie's points, along with various other images she's deconstructed and replied on. She covers a bit of photo manipulation, but that's after a photo's been taken. There's ways to manipulate the scene prior to taking an image, and she goes after that.

Folks should read her report with an open mind.

Images

[More_Cowbell]

PDF may be boring to you, but it has images, which are really necessary to this particular story

(whereas the link you provide does not.)

Re:Doctoring? Yes.

[Radon360]

And apparently you've never used a large clone brush with the source pointer overruning the modified result.

Here's a simple test. Set your clone brush to 100 pixels or so in size. Click the source point for cloning. Start cloning a 100 or so pixels away and drag the brush roughly inline with source point and clone brush centers. What happens? The pattern repeats itself at perfect intervals. Do this with a large, rectangular-shaped, hard-edge brush and you will get exactly the results in the doctored image.

You are correct that this is not an instance of a non-aligned clone process (i.e. clicking multiple points on the screen with the same clone source) in which it would introduce irregularities in the spacing. But the resulting image is quite evident of a clone brush "recloning" what it just did as it passed over the area it previously covered with the cloned area.

The excuse that this is an overzealous use of the dust/scratch removal is silly. If this guy were so concerned about the slight imperfection of dust on the orginial image, don't you think he'd notice that image had changed drastically after the application of this tool?

Proof that..

[Skadet]

experience is a necessary but insufficient condition for expertise. Look at the second picture, also by your good friend Hajj: http://zombietime.com/reuters_photo_fraud/

You know, the one with the cloned "missles" that were actually flares?

Oops.

He's done it before, you'd be blind not to think he did it again with this photo.