In 2009, computer programmer Adam Gurno was looking at photos from The New York Times Magazine online and thought one of the shots just didn't look right. The image of a room under construction seemed too symmetrical. Using graphics-editing program Macromedia Fireworks, Gurno created a mirror reflection of one half of the image and discovered that it matched the online photo perfectly. He notified the newspaper and the images were quickly removed.
Gurno told CNET that he just "got lucky" in managing to spot something more-trained eyes had missed. And while it's assumed that models are routinely Photoshopped in fashion magazines, in general it's not always easy to tell when an object or person has been removed from or inserted into a photo or when an image has otherwise been manipulated.
The ability to scrutinize images has become a topic of conversation since President Obama refused earlier this week to release any postmortem photos of Osama bin Laden, despite calls for him to provide them as evidence that the Al Qaeda leader is really dead. Meanwhile, online scammers wasted no time in distributing fake images of bin Laden with a bullet hole in his head in order to on malicious links.
The discovery of the original, unmodified photos of bin Laden on which the faked images were based provided immediate proof of the trickery. And though the shots might have fooled naive Web surfers, they were fairly transparent to experts.
Some telltale signs of manipulation are blurred areas; varied tones or shades; misplaced shadows and reflections; light coming from the wrong direction; and bits of a photo that repeat or create patterns, which could indicate that someone copied or "cloned" one part of the shot to cover or modify another area.
"I see fake photographs all the time on the Internet," said Richard Quindry, a photographer and Adobe Photoshop expert. "Usually you can tell just by looking. But to really know...you need to study it."
Quindry said he uses Photoshop to enlarge images and look for imperfections where two different images may have been joined or to see if pixels match up and to check that characteristics are uniform throughout the image. "I have techniques in Photoshop where I can magnify very minute differences and accentuate areas" where there might be anomalies, he said.
Ultimately, however, a conclusion using that technique is subjective, based on what the expert thinks he or she sees. For a more scientific reading, image-analysis software can be used.
Neal Krawetz, a computer forensics researcher, used a technique called Error Level Analysis to analyze a Victoria's Secret ad on The Hacker Factor Blog and found that the model's skin tone had been changed, her bust expanded, her eye color tweaked, her teeth brightened, her arm smoothed, her dress modified--indeed, her entire body had been dropped on top of the background, and a handbag she'd originally been carrying had been removed, fairly sloppily (the straps were still shown in her grasp, and the bricks behind the purse were bigger than all the others).
Error Level Analysis checks for differences in quality level throughout the image. Photos in the JPEG file format lose quality each time they're resaved, so additions to the image or changes will have a higher quality level than older areas that have been resaved more times, according to this FAQ. The site also has an online tool where you can insert an image URL for analysis. The site will spit back the image represented in dark and light areas, with the brighter areas indicating higher quality sections that would appear to have been resaved fewer times. If parts of the image are from different source files, they should stand out as a different color.
"If a picture has been resaved many times, or if a picture is of very low jpeg quality, it is quite difficult to determine accurately whether it has been digitally altered," the Error Level Analysis FAQ says. "Likewise, if all parts of a picture have been saved the same amount of times, it will not pick up, either."
A JPEG-file decoding utility, called JPEGsnoop, also lets people see if an image has been altered with Photoshop or other editing software.
But Photoshop and other tools are limited in their usefulness for this type of work, said Hany Farid, a computer science professor at Dartmouth whose team built customized forensics software for image analysis.
"We think about how people alter photos, what a forger does to create a compelling fake: adding shadows and reflections; removing something or adding it; or airbrushing--they all change some underlying statistical or mathematical aspects of an image," he said. "There is no simple answer. It's not like CSI, where you pull up a photo, make a couple of mouse clicks and boom, you're done."
His team looks at pixels and uses complex mathematical algorithms to carefully examine things like lighting, shadows, vanishing points, geometric properties, and other correlations, he said, "pointing the software in the right direction and letting it do its analysis."
"When you find traces of tampering, they can often be very compelling," Farid said. "But when you don't, you're left with one of two conclusions: either the photo is real or it's a very good fake." (Farid has a sampling of historical photo tampering here, and you can read his paper on how deceived the human eye can be here [PDF].)
The postmortem bin Laden photos that are circulating are very low quality and very small, which makes them difficult to analyze because the statistical clues are gone from the files, he said.
Proving something is a fake is much easier than proving it's real, Farid said. This means that even if the Obama administration released real photographs of bin Laden, that wouldn't satisfy the die-hard critics.
"This is the nature of conspiracies," Farid said. "The facts are not relevant."