Retraction Note: Salient object detection using a covariance-based CNN model in low-contrast images

Nan Mu, Xin Xu, Xiaolong Zhang, Hong Zhang, Jing HuNeural 2024 · DOI · OpenAlex

Introduction

In a move to preserve the integrity of the scientific record, the journal Neural Computing and Applications has formally retracted the 2018 article titled "Salient object detection using a covariance-based CNN model in low-contrast images." This retraction, issued in May 2024, follows an extensive internal investigation into the editorial and peer-review processes associated with the original publication.

The Crossmark 'Check for updates' icon is the standard industry mechanism for signaling retractions and updates to published research.

The Integrity of the Scientific Record

The original 2018 study aimed to advance computer vision by improving salient object detection—the ability of an AI to identify the most noticeable objects in a scene—specifically within challenging, low-contrast environments. However, the focus of the current notice is not on the technical merits of the proposed covariance-based Convolutional Neural Network (CNN), but rather on the systemic failures that allowed the paper to be published.

The retraction highlights a significant vulnerability in modern academic publishing: the exploitation of "guest-edited issues." These special editions, often managed by external experts, can sometimes bypass the rigorous quality-control measures standard to a journal's primary editorial board. The investigation suggests that this paper entered the literature through such a compromised channel.

Findings of the Forensic Investigation

The publisher’s internal investigation identified several critical procedural failures that led to the loss of confidence in the study’s results and conclusions:

  • Compromised Editorial Handling: The investigation concluded that the editorial process did not meet the journal's established standards for rigor and independence.
  • Peer Review Manipulation: Evidence suggested that the peer-review process was compromised, calling into question the validity of the "independent" feedback the authors received.
  • Inappropriate Referencing: The paper was flagged for potential "citation stuffing"—the inclusion of irrelevant references to artificially boost the impact metrics of specific researchers or journals.
  • Scope Mismatch: The article was found to be outside the intended scope of both the journal and the specific guest-edited collection in which it appeared.

Implications and Authorial Dissent

A formal retraction is considered a "nuclear option" in academic publishing. It indicates that the procedural or ethical breaches are so severe that the Editor-in-Chief can no longer vouch for any portion of the findings.

Notably, the retraction note records that author Xin Xu, representing all co-authors, expressed disagreement with the decision. This highlights the tension that often arises during retractions; while authors may defend the technical validity of their work, publishers may prioritize the integrity of the process used to vet that work. When the process is found to be corrupt, the underlying data is often treated as unreliable by default.

Conclusion

As of May 21, 2024, the 2018 study is officially retracted. The findings regarding covariance-based CNNs for low-contrast images presented in this paper are no longer considered part of the verified scientific canon. This case serves as a stark reminder of the ongoing need for robust oversight in peer review and the development of better tools to detect compromised editorial chains before they impact the scientific record.

Summary updated

BibTeX
@inproceedings{mu2024retraction,
  author = {Nan Mu and Xin Xu and Xiaolong Zhang and Hong Zhang and Jing Hu},
  title = {{Retraction Note: Salient object detection using a covariance-based CNN model in low-contrast images}},
  booktitle = {Neural},
  year = {2024},
  doi = {10.1007/s00521-024-10029-0},
}
Figure