Estimating a Null Model of Scientific Image Reuse to Support Research Integrity Investigations

How surprising is a match between two scientific images?

Image similarities need a reference point: a repeated pattern might be rare, or it might be common in scientific imagery. This preprint develops a statistical baseline for estimating how often a feature could occur by chance.

What the study found

• The method estimates the rarity of ORB image features against a large collection of scientific images.
• It provides a null model to help interpret suspected reuse during research-integrity investigations.

How the study works

High-dimensional density estimation is applied to features from more than seven million images in the PubMed Open Access Subset. Examples examine how estimated rarity changes with image complexity.

Scope and limitations

• Rarity depends on the feature representation and reference collection.
• A low probability under this model is not, by itself, evidence of intent or a finding of misconduct.

Cite this work

Daniel E. Acuna, Ziyue Xiang (2020). Estimating a Null Model of Scientific Image Reuse to Support Research Integrity Investigations. arXiv:2003.00878. https://doi.org/10.48550/arXiv.2003.00878

@article{acuna2020nullmodel,
  title = {Estimating a Null Model of Scientific Image Reuse to Support Research Integrity Investigations},
  author = {Acuna, Daniel E. and Xiang, Ziyue},
  year = {2020},
  publication_date = {2020-02-22},
  journal = {arXiv:2003.00878},
  doi = {10.48550/arXiv.2003.00878},
  url = {https://arxiv.org/abs/2003.00878}
}