A new study investigates our capacity to identify digitally manipulated images, and while its conclusion – that we are not very good at identifying doctored photos – is predictable, it’s the type of ‘fakes’ that deceive us which are most interesting.
In the paper ‘Humans Are Easily Fooled by Digital Images’ [PDF], researchers Victor Schetinger, Manuel M. Oliveira, Roberto da Silva and Tiago J. Carvalho asked approximately 400 test-subjects to identify images which had been manipulated by one of three methods: erasing, copy-pasting and splicing.
Erasing (below left) proved to be the biggest challenge for users to detect: removing an element or person from a photo, if done adroitly, is the most successful method of deception, followed by copy-pasting (below middle), in which existing elements of an image are duplicated elsewhere in the image. The least successful method, and – perhaps not by coincidence – that which requires the most skill of digital manipulators, is ‘splicing’ (below right), wherein an element is transplanted from a completely different image into an existing image.
The database used by the researchers contains 177 images, 80 of which are unretouched and 90 manipulated. The users were exposed to the photos in the form of a website, and, since the scientific image samples on display are not the most interesting when compared to what users are expecting to see in the context of ‘image manipulation’, the experience was converted into game-style sections.
The study was surprised to find that the ‘big bluff’, where a very large section of the image was manipulated or replaced, was as effective as any of the more subtle techniques. It also found that ‘grunge’ factors such as simulated or matched lens flares and deliberate mis-exposure was likely to produce a successful ‘fake’. Less than 58% of the imagery used was successfully identified, even though the subjects knew that they were being asked to locate manipulated images.
The researchers identified the ‘edited areas’ where reality and fakery coincided, and invited the subjects to individuate these where they suspected photographic fakery. Tell-tale signs included missing shadows and reflections, mismatched lighting on superimposed elements and perspective distortions that occur when two different images are taken with lenses that have varying focal lengths.
Removing unwanted friends, ex-partners and ‘image’ clutter from photographs makes this particular topic of scientific pursuit recognisable to the casual web user, but the image-set databases that have been developed for researchers are intended to help forensic science to calculate the veracity of photographic evidence.
The paper cites the 2012 case of ex-flight attendant Dimitri De Angelis, who succeeded in defrauding $8.5mn (AUS) from 16 investors with the aid of digitally doctored images depicting him in the company of luminaries such as Bill Clinton, Vladimir Putin and George W. Bush, and celebrities such as David Hasselhoff and Sophie Monk. It’s the kind of malfeasance that has become common enough for the likes of business litigation lawyer George L. Paul to declare that “photographs, as evidence of reality, are dead”, and attorney Zachariah B. Parry to state [PDF]: ‘Any time lawyers, police officers, and others routinely involved in lawsuits predict that a photo could potentially be used as evidence, they should assure that their photos can be authenticated. As the law stands today, the court has no way of guaranteeing that one of these methods will be utilized. Only by adopting one or more of these methods will the primary purpose of the courts be achieved: that justice be served.’