A paper gets retracted in Cell because of image manipulation.
Someone needs to tell scientists how not to use PhotoShop when preparing their images. I used to do precisely that to new researchers at the institute I worked at. Here are Wilkins' Rules of Scientific Image Manipulation:
1. If you have a blemish on your image, leave it there. Removing a blemish is like excising a word you don't like in a quoted piece of text. You may very well think it is irrelevant, but it may turn out to be the indicator of some artifact in your technique, so removing it looks like you are hiding something. Aesthetics do not matter in scientific papers; honesty does.
2. If you take separate sections of, say, a Western, leave a river of white space between them to show this (like using "..." to indicate missing text). It's too easy to make them look, illicitly, like that is the raw output. Which leads to
3. Keep unretouched copies of all images with the original date and time stamp. You might set up a directory that only the equipment can write to, and which gets archived by the lab's backup software. This is like taking and dating notes in your lab book, and for the same reason - to prove priority and honesty.
4. If you change the contrast, colour range, or add false colour, say this in the caption, explicitly. If necessary, state exactly what you did there (as a kind of "Methods" section for that image). If you add several images together, as in a two or three colour FISH, note this unless it is so obviously common practice that your readers can be expected to know about that method.
5. Any contrast or colour changes you apply to one part of an image should be applied to the whole. Local "improvement" is a kind of fakery. If you must enhance one section of an image, mark it out in a box, so the reader can see what you have done.
6. Be careful of applying filters, such as Unsharp Mask, to images. This can both obscure detail and add nonexistent detail, for instance, to cell microscopy.
In general, treat images in the same manner as you treat data. You don't exclude data just because it fails to back up what you want it to, and neither should you use images that way. Alternatively, treat images like text quoted from someone else - any missing or modified bits need to be marked.
- Log in to post comments
Great guidelines, John.
Mike Rossner from the Journal of Cell Biology recently had an opinion piece in The Scientist about how they examine digital image files for authenticity. He notes therein that the image-screening process could have caught part of Woo-Suk Hwang's fraud. Let me know if you have trouble accessing the article since it may be subscription-only.
I think photojournalists and newspaper editors have explored a lot of this ground already, and it may be worthwhile to see what they have to say.
For instance, using Photoshop to adjust the colors and contrast to make details easier to see, or to compensate for bad lighting, is okay. Cropping a photo to focus on one part of the scene is (usually) okay. Altering the actual content of the photo (e.g., removing a distracting bystander in the middle of the action) is not okay.
Obviously, there's a lot of gray area (if a photo shows children at play under the watchful eyes of combat soldiers, is it okay to crop out the soldiers?), but it's a good start.
The newspaper where I worked had additional rules about getting the shot. Any grip-and-grin type shot was out. It was very unlikely that a shot made with direct flash would get published, though it was not an absolute bar.
The most interesting thing in my opinion was the distinction made between asking people to do something that they had not already been doing to take a shot (bad) and requesting that they repeat something that they had just done in order to allow the photographer to make the shot (OK).
I recall being sent with one of our older staff photographers to a fraternity and sorority Valentine's Day exercize at getting a Guinness World Record for the most people simultaneously kissing. The actual action was going to be over in one minute, making it a bit of a challenge. My colleague, of course, got to ride up in the handy cherry-picker for an aerial view of the event. I was stuck on the ground. The people were actually arranged in a "heart" outline", so I walked around the outside looking for something, anything, other than the absolute usual thing of one or more couples in lip-lock. It turned out that one couple was a very short sorority girl matched up with a very tall fraternity guy, and the girl brought along a milk crate to stand on to bring them closer together in height for the big moment. I didn't have to ask anybody to do anything over and I got the shot. My editor rejected it, thinking it too perfect: he thought it too likely that readers would think the picture was specifically staged. Other than the basic artificiality of the event itself, it wasn't.
Speaking of children playing and soldiers being around, another of my colleagues, Al Diaz, came back from his hometown of Miami and the riots then going on there and pinned up an 11x14" print in the darkroom. He used a wide-angle lens with the camera near the ground. In the foreground, there were two African-American boys of about 10 years of age smiling and mugging for the camera. About 10 feet behind them, there was a National Guardsman holding an M-16, and then receding into the distance one could see more Guardsmen spaced evenly along the palm-tree dotted median of the now-deserted highway. Smoke from a fire nearby drifted over the whole thing. It was a stunning image, though unsuited for publication because of the direct interaction of the children and the camera. The fourth wall is no less a barrier in print photojournalism.
I suggest that there's a point that's being missed here. Pick up a brand new digital camera and take a picture. Things like contrast, colours, etc. **have** been manipulated according to the standard profile that the manufacturer decided to set.
Take a picture in RAW and save it without touching anything, and (as I understand it) you're still not quite clean because if depends on the variety of RAW being used. In other words, unmanipulated RAW pics from different cameras may be visibly different.
Manipulation is manipulation whether it's done by the in-camera processor or via PhotoShop on your PC ... sorry, on your Mac. In other words, I'm not sure that the "unmanipulated photo" really exists.
This is not to discount your good advice, just to suggest that things aren't quite that straightforward.
Another way of helping this would be for journals to require untouched photos to be included as supplementary supporting material. The pretty touched-up version would be directly in the paper, but the original* images would be sent to the reviewers to check and would also be available for any else interested.
( * whatever "original" means in the digital age, as other commenters have noted )
--Simon
@Scott Belyea
That is a good point. I think you should always specify what equipment was in use if you want an image (especially a digital image) to be regarded as evidentiary.
The camera or equipment used should be in the materials and methods section, surely.
I think the idea of the unretouched files being supplementary data in an open source format (TIFF, obviously) at the journal's site is a very good idea. It should also be available on request from the institute or lab.
Good advice. I think we need to remember that these potential problems predate digital methods. Darkroom-processing introduces the opportunity for all sorts of artifacts, both intentional and otherwise. It is just more difficult and requires specialized skills and substantial training.
My wife worked as a digital-imaging specialist for a veterinary school for a time and prepared lots of figures for research papers. She had very strict guidelines to follow. Some of her clients were frustrated because they wanted their gels to appear a certain way that required inappropriate cutting, pasting, sharpening, and so on. Thankfully, most understood and accepted the restrictions and worked within them. I wonder if they would have been more comfortable with the less-than-perfect gels and not so ready to look for ways to 'clean up' their results if digital manipulation had not been possible. Of course, the djinn is out of the bottle now so guidelines like the ones you outlined above are good to have.