Manipulated images fool detection systems

Using advanced image restoration techniques, miscreants can now fool detection systems into thinking manipulated JPEG images are the real thing. That’s the unsettling finding of research done by four engineers from Gipsa Lab and Peking University, who won the Best Paper Award at a June conference.
Existing detection systems work by spotting statistical “footprints” left by JPEG compression when one image is inserted into another. The engineers were able to evade these systems by aggregating their different action models to develop a restoration model that can pass undetected.
Thus the cat-and-mouse game between fraudsters and detection systems continues. But it’s not just child’s play—in March North Korea’s doctored photos of its supposed military hovercraft made news headlines around the world.

Contact: francois.cayre@gipsa-lab.grenoble-inp.fr

Facebook

Twitter