A novel blocked based U-Net model for image steganography

This article introduces a novel approach for information hiding by leveraging the potential of advanced deep learning based model. Using U-Net architecture and block-based input, this study aims to improve the capacity and visual quality of techniques to hide information. The proposed novel U-Net model offers a more compact network design that can be easily utilized for information hiding apart from information extraction tasks. Initially, during first step of information hiding phase, the cover as well as secret images are being split into non-overlapping chunks of blocks. Then the secret image blocks are compared with the cover image blocks and the best Structural Similarity Index Measure (SSIM) block is chosen for embedding. Afterwords secret image blocks are embedded within the cover image. During the process of extraction, the hidden image’s features are being extracted, leading to the recovery of the concealed secret image. Adopting the U-Net structure ensures intricate pattern would be preserved from both the carrier image and the secret image. Simulation experiments demonstrate a significant enhancement in capacity compared to traditional methods, along with impressive visual results. The proposed approach reported a higher value of SSIM (0.9829) and PSNR (40.5166) for stego image whereas 0.9904 and 40.6872 for Re-secret image respectively.