Distributed Anti-Plagiarism Checker for Biomedical Images Based on Sensor Noise

The increasing number of scientific papers reporting false or stolen data calls for the needs of new tools able to automatically detect plagiarism or unfaithful ownerships. This problem is particularly actual for the health sciences, as the number of biomedical images that are stolen or manipulated and, then, published in scientific papers is becoming higher and higher [1]. In this paper we present an automatic anti-plagiarism checker that relies on the concept of Pixel Non-Uniformity (PNU) noise. This is the characteristic noise left by source sensors of devices like digital cameras, electron microscopes or Magnetic Resonance Imaging (MRI) to define a sort of fingerprint for these devices. The intended use of our system requires two steps. In a first step and on a voluntary base, the researchers register to the system their imaging devices by providing a training set of images. These will be used to extract the device fingerprint called Reference Pattern (RP). In a second step, the system will periodically scan a set of known scientific digital libraries (most publishers offer on-line access to their papers) downloading the new papers and extracting all the images herein contained. The output produced by a specialized filter on such images will enable the system to compare the Residual Noise (RN) with all the enrolled device patterns, allowing the identification of the device that captured the image. Given the huge amount of papers and images to process, our system has been implemented as a distributed application running on top of the Spark cluster engine