Scientific Publication Condemns Photo-Manipulation

valdean writes "According to a recent article in the New York Times (registration and short biography required) scientific journals have begun to respond to a growing problem of photo-manipulation in submitted manuscripts. At The Journal of Cell Biology, a test developed in 2002 revealed that 25 percent of all accepted manuscripts had one or more illustrations that were manipulated to the point of violating the journal's guidelines. Examples included the duplication of images for re-presentation as a control experiment, making pictures prettier with the clone stamp tool, use of the contrast tool to hide data, and merging portions of several images so that they appear to be a single image. How were many of these scientists caught? They submitted layered Adobe Photoshop files that showed exactly what they had done."

Comments

Re:Question

[faloi]

It depends on the skills of the person manipulating the image, and the image quality. The higher resolution the picture, the easier it is to zoom in and spot anomalies. If the picture is craptastic to begin with, it's harder to see the differences (tough to tell if the blurring is a result of someone mucking with the picture, or just poor quality).

The biggest roadblock to telling whether an image is real or not is time, in my opinion. If you generally trust the person providing the photos and they're not too unbelievable, you probably won't spend time trying to figure out whether it's reliable or not.

Re:Question

[whorfin]

In this case, they had the source, which tells me that the scientists that got caught weren't exactly the sharpest spoons in the drawer.

Here's a prior slashdot posting about mathematical techniques to identify photo manipulation. And another article detailing some techniques.

Not layered images

[dmoore]

Unless I'm mistaken, the article never states that the scientists submitted layered Photoshop images that revealed their misdeeds. I find it very unlikely that a journal would ever accept an image in photoshop format -- they usually want press-ready formats like PDF, EPS, or JPG.

From the article, it sounds like the editors just fool around with brightness and contrast of submitted images, and that often reveals the discontinuities from an edit. However, the specifics are not in the article, so don't jump to conclusions.

Re:Not layered images

[rco3]

Scientific Journals, like the various AGU publications, typically prefer photographs in .jpg, false-color data in .tiff, and charts, graphs, and other line-art suitable images as .eps. .Png is acceptable for electronic versions. Photoshop .psd's are not acceptable, and .ai files are acceptable for review / draft stages only. In fact, the IEEE won't accept .jpg files at all. This may differ from what Redbook, Car and Driver, and Rolling Stone want; I've never been published in one of those. But considering that the images under discussion are scientific images, I'd say that .eps files are most definitely on the preferred format list.

Re:Question

[merlin_jim]

Others have made good replies to this but thought I'd add my $0.02

Generally, image manipulation will leave a signature of some sort on the file - do a fourier transform (view the image as frequency data as opposed to spatial) and you can see some of them pretty clearly. They generally show up as very low or very high frequency noise distributed more or less uniformly around the origin. Then there's edge detection; most computer-based photo manipulation creates or erodes edges and a basic edge detection algorithm will show the problem to most human observers.

As mentioned by others, a low quality original can make it much harder to detect manipulation.

Allowed Image Manipulations

[wsherman]

From the summary:
...violating the journal's guidelines. Examples included ... use of the contrast tool to hide data...

The actual article is a bit more nuanced:

To prohibit such manipulations, Dr. Rossner and Dr. Mellman published guidelines saying, in effect, that nothing should be done to any part of an illustration that did not affect all other parts equally. In other words, it is all right to adjust the brightness or color balance of the whole photo, but not to obscure, move or introduce an element.

If a researcher manipulates only part of an image then the researcher is implicitly admitting that there was something in the image that they chose to ignore. On the other hand, if a researcher changes the contrast of the whole image to make it easier to see the patterns they are drawing their conclusions from, then they can always claim they really didn't notice the other stuff. Essentially, researchers have to avoid doing things that prove that they were deliberately dishonest.