You may have already heard about the Chinese antelope that weren’t: This, from WSJ’s Jane Spencer and Juliet Ye:
Earlier this week, Xinhua, China’s state-run news agency, issued an unusual public apology for publishing a doctored photograph of Tibetan wildlife frolicking near a high-speed train.
The deception — uncovered by Chinese Internet users who sniffed out a Photoshop scam in the award-winning picture — has brought on a big debate about media ethics, China’s troubled relationship with Tibet, and how pregnant antelope react to noise.
The photographer and editor involved have since resigned. But this took two years to out; you look at the photo now and you just know that it’s not real. And this, of course, is not the first time photos have been doctored by news organisations that should know better (there’s Reutersgate, as it’s sometimes called, when Lebanese freelance photographer Adnan Hajj was caught allegedly duplicating flares, buildings and plumes of smoke for Reuters. More on photoshopping at Wikipedia). How can we avoid this?
One option is to be a bit more discerning about the pictures we see, whatever their provenance. Another is to turn to technology. Academics Jessica Fridrich, David Soukal and Jan Lukáš looked at
detection of a special type of digital forgery – the copy-move attack in which a part of the image is copied and pasted somewhere else in the image with the intent to cover an important image feature.
Their paper [PDF] investigated the problem of detecting the copy-move forgery and described what they called “an efficient and reliable detection method”. John Graham-Cumming, best known for his work on Bayesian spam filters, made it a reality with an algorithm that implements automatic detection of image alteration using copy/paste. (OK, he did it because he wanted to win money in a spot the ball competition, but it’s still good work.)
A guy called John Wiseman has made a few modifications to the code so it works faster, and has shown how it works well in detecting the alterations in Adnan’s photos:
The blue and red bits are where there are duplicated pixels. The code didn’t work very well on the Chinese antelope picture because that involves splicing two pictures together more than copy/move. But it’s worth a look.
Is this going to bring to an end Photoshopping? Probably not. But it might make us more skeptical, and if tools like this are readily available, more likely to run suspect photos through the wringer until we’re sure that what we see actually happened like that.