TRAIN THE LATENT, NOT THE IMAGE: JOINT IMAGE COMPRESSION AND STEGANOGRAPHY
Keywords: steganography, image compression, semi-amotorize, adversatial learning
Abstract: Image steganography is the process of hiding secret information in an image through imperceptible changes. Most of recent works achieve message in the image by modifying the pixels of image itself. However, those images with hidden messages are not robust to compression such as JPEG, which is used almost everywhere. In order to achieve the ability to compress the image while still having the ability to carry the message, we propose an innovative optimization method which leverages a semi-amortized approach to directly manipulate latent space data for the joint optimization of image compression and steganography. In the compression module, we investigate two of the most popular models in learned image compression with different pre-trained quality: the hyperprior model and the ELIC model. For the steganography module, our method employs the pre-trained fixed neural network steganography (FNNS) model. We compare our method with two state-of-the-art methods such as FNNS-JPEG and LISO-JPEG, achieving significant image compression while maintaining high fidelity and ensuring the accuracy of content upon decoding. The results demonstrate the effectiveness and superiority of our approach.
Primary Area: applications to computer vision, audio, language, and other modalities
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/2025/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: 6559
Loading