Home > Archives > December 2005 >

How To

How to Tell If a Digital Image Has Been Altered

By Simson Garfinkel

E-mail this article  |  Printer friendly

Advertisers

Digital photography, Photoshop and synthetic computer graphics have made image trickery both easier to commit and harder to detect. But it is still possible to catch tampering—provided you have a good eye and the right tools.

Physical impossibility is a good giveaway. For example, a current advertisement for a four-wheel-drive sedan that shows lots of two-wheel-drive vehicles up on their back wheels ("Why pay for four wheels if you’re only using two of them?") is clearly faked because the shadows on the pavement don’t match the cars up in the air. So sometimes, detecting a forged image simply requires looking closely. In another recent high-profile example, the crowds in photos released by the Bush reelection campaign in October 2004 had been digitally enhanced with extra faces. Careful observers found that some of the faces were present more than once in the same crowd shot.

Another visual clue: Look pixel-by-pixel at a digital image and you might be able to see the sharp lines that result when one image is pasted on top of another.

There are also some clues unique to the digital world. For example, the vast majority of today’s cameras take .jpeg-format digital pictures at predetermined sizes such as 640x480 or 1024x768. If you come across a .jpeg that’s an odd size—say, 500 pixels square—then you know that part of the original image has been cropped.

Researchers at Dartmouth College have developed sophisticated algorithms for detecting manipulation of digital images in general and .jpegs in particular. Because .jpeg is a compressed file format, any tampering with an image from a digital camera usually results in the .jpeg image being twice-compressed: once by the camera, and once by Photoshop. This double-compression leaves tell-tale artifacts in the image that can be detected through the use of advanced algorithms. Today this kind of technology is still in the research laboratory, but if you have a copy of Matlab you can download the algorithms and try them out yourself. Also on the website is an interesting collection of digitally tampered magazine covers.