Fraud-busting program hunts for doctored pictures.
The editors of scientific journals could catch fraudulent images by using computer tools similar to those being developed for law enforcement and photojournalism, say computer scientists.
Recent fraud at the lab of stem-cell researcher Woo Suk Hwang (Nature 439, 122–123; 200610.1038/439122a) highlights the ease with which scientists can cook up fake images. In the South Korean team's 2005 Science paper, for example, two photographs in the same figure were found to be partial duplications.
The Hwang furore is encouraging journal editors to seek ways of detecting suspicious images before they are ever published. Such screening could identify outright fraud, as well as a far more pervasive practice in which biologists use programs such as Photoshop to tweak and smarten scientific images before publication (see Nature 434, 952–953; 2005).
At least one group of journals, published by the Rockefeller University Press (RUP), already carries out such image forensics by eye. Mike Rossner, managing editor of the Journal of Cell Biology in New York City, has trained a production editor to enlarge and scrutinize images for obvious but illegitimate changes, such as bands erased from a gel or cells slipped into a microscope image. This production editor checks manuscripts accepted for publication at the Journal of Cell Biology and two other RUP journals.
But this is time consuming, and computer scientists say that computer algorithms could automatically scan digital images and ferret out signs of manipulation. The development of such systems, prompted by the explosive spread of digital cameras and imaging programs, is also of interest to lawyers and police, who want to check for tampered crime-scene images, as well as to news organizations keen to detect faked photos.
Partly funded by the FBI, Hany Farid at Dartmouth College in Hanover, New Hampshire, and his colleagues have designed a suite of ten mathematical techniques to scan images for the hallmarks of manipulation. For example, one algorithm searches for small clusters of identical pixels in an image, and so might reveal where an area of background has been copied and pasted over a blemish.
A second algorithm can identify whether part of an image has been expanded, perhaps to splice two photos seamlessly together. To do this, a program such as Photoshop generates new pixels by averaging the characteristics of the neighbouring ones — leaving a giveaway signature in the image.
Farid is currently converting his algorithms into a user-friendly form that can be attached to ImageJ, a free image-processing program distributed by the US National Institutes of Health. He is also consulting Adobe, the company that makes Photoshop, about whether the algorithms could be packaged together as a plug-in for the program, for use by different industries. This could help journal editors and reviewers, but also lab heads wanting to check the work of their team.
Rossner says he plans to trial Farid's system when it is complete. Editors at Nature are also consulting researchers about automatic tools for detecting image manipulation.
But there are drawbacks. Such a system could act only as a first line of policing to flag up suspicious images. And editors and technophiles alike agree that anyone determined to fabricate or alter images will be able to fool the forensics software — perhaps by using the very same detection algorithms to learn how. Eric Postma is a computer scientist who devises software to detect fine-art fraud at the University of Maastricht in the Netherlands. “It's always a race between two sides,” he says.
Related links in Nature Research
Verdict: Hwang's human stem cells were all fakes
Special Report: Taking on the cheats
Image manipulation: CSI: cell biology
Related external links
Rights and permissions
About this article
Cite this article
Pearson, H. Forensic software traces tweaks to images. Nature 439, 520–521 (2006). https://doi.org/10.1038/439520b
This article is cited by
Generalizing on Best Practices in Image Processing: A Model for Promoting Research Integrity
Science and Engineering Ethics (2010)
Avoiding Twisted Pixels: Ethical Guidelines for the Appropriate Use and Manipulation of Scientific Digital Images
Science and Engineering Ethics (2010)
Annals of Biomedical Engineering (2008)
Your own desktop crime lab
Nature Medicine (2006)