Image Steganography Scheme Using Dilated Convolutional Network

• Published in: 2021 12th International Conference on Information and Communication Systems (ICICS) pp. 305 - 309
• Main Authors: Kich, Ismail, Ameur, El Bachir, Taouil, Youssef, Benhfid, Amine
• Format: Conference Proceeding
• Language: English
• Published: IEEE 24.05.2021
• Subjects: CNN; Color; Communication systems; Deep learning; Deep Neural Network; Encoder-Decoder; Generative adversarial networks; Image steganography; Information security; Receivers; Security; Visualization
• ISSN: 2573-3346
• DOI: 10.1109/ICICS52457.2021.9464546

Nowadays, Convolutional Neural Networks (CNN) have allowed us to solve many problems, difficult to solve with classical methods, in different fields of applications. The use of this technique in the field of modern steganography has improved the performance of steganographic schemes in terms of concealability and invisibility. In this article, we propose a system based on CNN in order to hide a color image in another color image of the same size. The proposed system is based on Auto-Encoder network and U-net architecture. The network is subdivided into two sub-networks, the first is for the concealment of the image secret by the sender, the second is for its extraction by the receiver. The network is end to end trained to ensure the integrity of the concealment and extraction process. The tests were performed on challenging images dataset publicly available, such as ImageNet, LFW, PASCAL-VOC12. The results show that the proposed steganographic scheme can hide a color image in another one of the same sizes, i.e. a capacity of 24 bpp, with acceptable PSNR and SSIM values compared to other previous work.