UNCOVER researcher wins best paper award at WIFS2023 for his work on image steganalysis

UNCOVER researcher wins best paper award at WIFS2023 for his work on image steganalysis

Rony Abecidan, a researcher from the Centre National De La Recherche Scientifique (CNRS), partner of UNCOVER project, located in Lille, France, has received the best paper award at the IEEE International Workshop on Information Forensics and Security (WIFS) in Nuremberg, Germany, on December 4-7, 2023.

His paper, titled “Leveraging Data Geometry to Mitigate CSM in Steganalysis“, was co-authored by Vincent Itier, Jérémie Boulanger, Patrick Bas, and Tomáš Pevný, and presented a novel method to improve the performance of image steganalysis models in realistic scenarios.

Image steganalysis is the technique of detecting and extracting hidden messages in images, which can be used for various purposes such as digital forensics, cybersecurity, and privacy. However, image steganalysis models often suffer from a problem known as Cover Source Mismatch (CSM), which occurs when the test images come from different sources or processing pipelines than the training images. This leads to a significant drop in the accuracy and reliability of the models.

To address this challenge, Abecidan and his colleagues proposed a geometry-based optimization strategy that enables the selection or derivation of customized training datasets that match the target test images as closely as possible. Their method is based on a geometrical metric that measures the similarity between the subspaces spanned by the features extracted from the images. By using this metric, they were able to identify or construct very relevant training datasets for a large variety of targets minimizing the operational regret, which is the difference of performance between the optimal model and the model used in practice.

The paper also presented experimental results that demonstrated the effectiveness and robustness of their method, which outperformed the traditional methods that rely on statistical or pixel-level analysis.

The paper was selected as the best paper among various papers submitted to the WIFS2023, which is the flagship workshop of the IEEE Signal Processing Society for information forensics and security. The workshop covers a wide range of topics related to the theory and practice of information security, such as cryptography, watermarking, biometrics, multimedia forensics, and adversarial learning.

The best paper award is a prestigious recognition of the high-quality and original research conducted by Abecidan and his collaborators, as well as a testament to the excellence and impact of the UNCOVER project.

For more information about the paper and the workshop, please visit the official website: https://ieeexplore.ieee.org/document/10374944

You can also read the paper here: https://hal.science/hal-04229257v1