• 
• 
• 

 1 of 3

Gallery

How experts identify the fakes

Posted on 13 May 2010 at 15:43

We look at the techniques and software used to identify Photoshop fakery

There are plenty of Photoshop fakes out there, but how do you tell the real deal from the ruthlessly altered picture? Software can help, but your best bet is still a sharp eye.

A new person

In this famous photo of former work and pensions secretary James Purnell, the man in question was running too late to meet his colleagues and make it into the final shot.

Not to worry: the other people in the photo obligingly stood to the side in order to tell a tiny fib and allow Purnell to look as though his watch worked properly. A cursory inspection of the photo reveals that he was added afterwards, though.

According to expert Hany Farid, detecting where a person has been added to a shot is all about the lighting. Purnell stands on the far right of the photo, apparently evenly lit from the left-hand side.

But his colleagues appear to be lit from above and to the right. We might also note that Purnell is standing uncomfortably close to Kevin Corscadden. An even closer inspection reveals that the cut-out around Purnell’s hair – where he’s been removed from his original background – is incomplete. A true Photoshop disaster.

Error level analysis

For an insight into the technical work behind detecting faked images, visit Error level analysis, which claims to help detect elements that have been added to a photo.

Every time you save an image as a JPG, you lose a tiny amount of detail. Over time these small losses add up, and eventually it should be possible to tell if two images have been blended into one, if one picture has been saved more times than another.

There are all kinds of reasons the car-crash of a composition below would lose you your job at a picture desk, and it’s clear that the hurriedly added person wasn’t standing there when the shutter release was pushed. But a little technical analysis can help.

Error level analysis processes an image by saving it again, then compares the old image to the newly saved one and determines how much detail has been lost and from where. The key is the brightness of different parts of the image. If an image is returned and is uniformly bright (or dark), it suggests that little or no manipulation has happened. In the first, person-less print there are only a few highlights – largely on the finely detailed twigs on the right-hand side.

This is largely theoretical: many of the images we ran through the tool suggested Photoshop tampering when there was none, while other images came back looking relatively clean when in fact new elements had been introduced, so it’s clear you couldn’t build a criminal case around it. However, this image demonstrates the potential.

Pay attention to the dark-blue squares covering the jacket – this chunky, square texture isn’t found elsewhere in the image, and is indicative of a part of the image that has been saved with a different level of compression than other parts of the photo.

Author: Dave Stevenson