Exposing digital forgeries in scientific images

A recent case of scientific fraud, involving manipulated images in a high-profile scientific publication, has sent shock- waves through the scientific community. By some measures, however, this case is not isolated - in at least one journal, it is estimated that as many as 20% of accepted manuscripts contain figures with inappropriate manipulations, and 1% with fraudulent manipulations. Several scientific editors are considering putting safeguards in place to help reduce these numbers. While sensible policy and awareness are certainly important, there is likely to be a need for computational techniques that automatically detect common forms of tampering. We describe three such techniques for detecting traces of tampering in scientific images. Specifically, image segmentation techniques are employed to detect image deletion, "healing", and duplication.