As an editor, what measures can I implement to prevent image/photo manipulation?

Question

There's been a case of alleged data manipulation in a recent Nano Letters paper (initial report here; now-retracted paper):

    

You can surely make your own opinion on whether the images have been digitally modified, given the above snapshots (look for gray rectangles around the rods); I was careful to write “alleged” because the paper’s PI threatened legal action against the blogger who reported the issue.

---

Now, this case lead to a large number of comments here and there about whose responsibility it was to catch this issue (reviewers or editor). I tend to agree that the reviewer should probably have caught it, but this lead me to wonder:

As editor of a scholarly journal, what measures can I implement to prevent image/photo manipulation? And what about data manipulation? Organic Letters has made the news two months ago when they hired a in-house data analyst. Is that the way to go? Are there other measures one could take to reduce the threat of image and data manipulation?

Answer 1

I think a search term for this type of manipulation detection is image forensics. Matthias Kirchner: Notes on Digital Image Forensics and Counter-Forensics may be a starting point.

What can you do?

I think a first step is to communicate clearly what image manipulations* are acceptable and which are not:

• is it acceptable to adjust brightness and contrast?
• correct unequal illumination?
• whitelight correction based not on a whitelight measurement but on parts of the image?
• digital sharpening or other enhancement filters?
• where's the border between an inset picture and a fraudulent manipulation?

and so on.

Who should detect this?

I agree that the reviewer should have commented and asked about the pictures above, however my experience as reviewer is that I often get pictures in ridiculously low resolution (I comment on that). I'm not sure, but I think that the publishers may ask for high resolution pictures when they spot such problems in the production process. However, that would mean that the reviewers may not have seen the actual picture that is used for printing.

---

* manipulation in the data analysis sense: calculations that change the information content (e.g. enhance contrast), and cannot be easily undone (as opposed to a transformation where the back transformation is easily possible, e.g. rotation), not in the sense of fraud.

Answer 2

There exist softwares dedicated to this. Journal of Cell Biology seems to have been a precursor in this in 2002.

A first simple check suggested in this paper is to open images in a image manipulation software and change "controls" (I guess: contrast, luminance, etc.) The problematic regions can then appear clearly.

I would say that, like plagiarism, this is the publisher to check or provide software to the editorial board, as this is technical and not scientific assessment.

Answer 3

It seems to me the only possibility is to subject all or particularly important images to an analysis detecting variation in noise levels (see examples here; service unfortunately closed). I personally do not know if such an analysis could be automated in an electronic submission system or be used as a tool by the editors. How much effort should be spent looking for fraud will obviously depend on assessments of, for example, how critical images are for the publication(s), the likelihood for fraud (random tests?) and the cost in terms of time and money for doing the testing.

It seems to me there would be much use to have a tool to do such an analysis for all submissions. The problem is of course that the noise level detection also identifies all kinds of manipulation and so it would seem reasonable to ask authors to provide a very detailed account of what has been done to each image so that the analysis can be set in a perspective.

Answer 4

"Kill the chicken, and make the monkey watch"

Institute an editorial policy by which all convicted offenders are banned for life from the publication in question. Convince other publications to share such info, and act in the same way after definitive proof is presented. No second chances.

Publicize these events far and wide.

That will probably take care of 60-70% of the problem, at least the blatant cases such as the one above.

Unfortunately few publishers have the conviction to do that.

EDIT: To answer your question: "Who do you ban among N coauthors?" To first approximation, all of them as co-authorship implies shared responsibility. However, that can be more accurately determined on a per-case basis as the result of detailed investigation.

Detecting instances fraud is trivial if crowdsourced. Blatant image manipulation as the one shown above would eventually have been noticed by a reader of that paper. Same applies to other similar kinds of fraud. Relying on a single overburdened editor and a couple of bored referees for that task makes it much more difficult. Hiring staff to essentially redo part of the research reported in manuscripts submitted to the journal is just laughable.

The point is to demand ethical standards as a publisher, and raise the stakes so high that the penalty of getting caught, guaranteed loss of professional reputation and possibly employment, offsets any gain from publishing a single or a series of papers.

Relying on automatic detection schemes is inherently unreliable. Smart people will always find creative ways to cheat more effectively. To illustrate my point, consider the unending arms race between virus writers and antivirus software companies.

To conclude, and since you seem to be quite green, I suggest you give this a thorough reading: https://en.wikipedia.org/wiki/Schon_scandal

It won't take many Schönen to drive the point home...