FStega: Fourier Neural Operators for printer-proof steganography
Keywords: steganography, watermarking, fourier operators, deep neural networks
TL;DR: In this paper, we introduce a method using Fourier Neural Operator to embed bitstrings in printed images, enhancing data recovery while maintaining image quality, achieving high accuracy across various conditions and image sizes.
Abstract: Hiding and extracting a message in printed images is challenging when considering the trade-off between accuracy in recovering data and the perceptual quality of the generated images. It is especially a big issue to hide data in such a way that it is almost invisibly embedded. In this paper, we propose a method based on Fourier Neural Operator to embed bitstrings in images. The method is able to learn critical frequencies from the image and the message to improve the decoding process of the hidden data. In order to enhance the information recovery capabilities of the printed image we create an improved noise simulation process and a decoder composed of several convolutional layers combined with a vision transformer to obtain a decoding method more robust to the noise introduced in the encoding
image when it is printed and acquired by an optical sensor. Experimental evaluations demonstrate the ability to properly recover the message encoded in wild printed pictures with an accuracy of 100% (with 3 different image sizes) acquired with several lightning and perspective conditions.
Primary Area: general machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8005
Loading