Abstract

The quality and authenticity of images is essential for data presentation, especially in the life sciences. Questionable images may often be a first indicator for questionable results, too. Therefore, a tool that uses mathematical methods to detect suspicious images in large image archives can be a helpful instrument to improve quality assurance in publications. As a first step towards a systematic screening tool, especially for journal editors and other staff members who are responsible for quality assurance, such as laboratory supervisors, we propose a basic classification of image manipulation. Based on this classification, we developed and explored some simple algorithms to detect copied areas in images. Using an artificial image and two examples of previously published modified images, we apply quantitative methods such as pixel-wise comparison, a nearest neighbor and a variance algorithm to detect copied-and-pasted areas or duplicated images. We show that our algorithms are able to detect some simple types of image alteration, such as copying and pasting background areas. The variance algorithm detects not only identical, but also very similar areas that differ only by brightness. Further types could, in principle, be implemented in a standardized scanning routine. We detected the copied areas in a proven case of image manipulation in Germany and showed the similarity of two images in a retracted paper from the Kato labs, which has been widely discussed on sites such as pubpeer and retraction watch.