Images courtesy of Hany Farid
An original (left) and doctored (center) image. The image on the right was generated by two algorithms for detecting digital tampering.
Back when “photography” meant film and negatives, few people could convincingly retouch a shot of President Bush giving his State of the Union address or craft a realistic photo of John Kerry at a war-protest rally next to Jane Fonda. Now that such possibilities are just a mouse click away, computer scientist Hany Farid of Dartmouth College is developing mathematical tests to identify altered, potentially misleading images.
Each digital image contains a grid of pixels that encodes information about color and brightness; collectively, the information in an image tends to form a distinctive statistical pattern. Farid and his students have so far developed six algorithms that detect disruptions in that pattern, including those caused by differences in the resolution or graininess of separate original images. If part of a photograph is rotated or scaled to match (as was done in the doctored Kerry-Fonda snapshot) such manipulations will show up as well.
Farid is sharing his mathematical code with federal law-enforcement agencies. He anticipates that the technology will eventually be used in courts to validate crime-scene photos or other digital evidence. Pixel sleuthing, Farid recognizes, will make a forger’s job more difficult but not impossible: “No matter what we develop to detect tampering, there are ways around our detection scheme.”