Gauging photo fakery

02 December 2011
Posted by Chris Smith.

A technique to quantify the perceptual impact of photo-manipulation onPhoto manipulation fashion shots has been developed by US scientists.

On a daily basis, billboards, magazines and websites dish up a deluge of wrinkle and blemish-free, impossibly-long-legged, large-breasted and extraordinarily thin fashion "icons" and celebrities.

But this type of creative artifice could be distorting peoples' perceptions of social norms and may be linked to eating disorders, anxiety states and body dysmorphic disorders amongst vulnerable individuals who find expectations of themselves impossible to live up to.

So great is the concern that some countries, including the UK, France and Norway, are even considering introducing legislation that will force photo-retouchers to show their hand and be transparent about their use of "Photoshop" techniques.

Merely marking an image as manipulated or not, though, is of limited value because such a label doesn't distinguish between standard techniques - like cropping or colour balancing - and more dramatic modifications of physique, skin turgor and appearance.

Now, writing in PNAS, two Dartmouth College, New Hampshire-based computer scientists, Eric Kee and Hany Farid, have developed a mathematical algorithm that can accurately gauge the perceptual extent to which a picture has been modified.

The duo began by trawling the web to locate 468 before and after images of retouched individuals. They then engineered a formula to compare the two sets of pictures from both geometric - meaning changes to body shape - and photometric - changes to colour, tone and blemishes - perspectives.

They then trained their model by showing the images to 390 online volunteer observers sourced internationally using a crowdsourcing platform called Mechanical Turk. These individuals rated the degree of manipulation visible in each of the images on a scale of 1-5.

The results showed that the manipulation scores returned by the mathematical model are a robust reflection of what a human sees. The researchers therefore suggest that their model could be employed industry-wide to provide viewers with an objective measure of the degree to which any image reflects reality.

They also acknowledge, however, that if such a system is introduced crafty artists may well indulge in a game of graphical cat and mouse to subvert the system...

Add a comment