Scientific Image Tampering Detection Based On Noise Inconsistencies: A Method And Datasets

Can image noise help identify manipulated scientific figures?

Scientific images have different properties from everyday photographs. This preprint develops a detector tailored to scientific imagery and tests whether inconsistencies in image noise can reveal manipulated regions.

What the study found

• The method is trained and evaluated using manipulated western blot and microscopy images.
• On the reported benchmarks, it outperformed the general-purpose image-tampering methods used for comparison.

How the study works

The work constructs scientific-image manipulation datasets and learns to detect noise inconsistencies associated with edits.

Scope and limitations

• Benchmark performance on these image types does not establish performance on every scientific imaging modality.
• A detector output is a signal to inspect the source material, not a determination of research misconduct.

Cite this work

Ziyue Xiang, Daniel E. Acuna (2020). Scientific Image Tampering Detection Based On Noise Inconsistencies: A Method And Datasets. arXiv:2001.07799. https://doi.org/10.48550/arXiv.2001.07799

@article{xiang2020tampering,
  title = {Scientific Image Tampering Detection Based On Noise Inconsistencies: A Method And Datasets},
  author = {Xiang, Ziyue and Acuna, Daniel E.},
  year = {2020},
  publication_date = {2020-01-21},
  journal = {arXiv:2001.07799},
  doi = {10.48550/arXiv.2001.07799},
  url = {https://arxiv.org/abs/2001.07799}
}