Efficient U-shape invertible neural network for large-capacity image steganography

Abstract

Image steganography ensures covert communication by hiding secret information within cover images. The existing low-capacity steganography methods achieve satisfactory performances when hiding limited binary information within a cover image. However, it is still a challenge to recover high-quality revealed secret images from highly secure stego images with limited computational cost for large-capacity image steganography. This paper proposes an Efficient U-shape Invertible Neural Network (EUIN-Net) for large-capacity image steganography. Due to the gradual fusion and separation properties of the U-shape invertible mechanism, our EUIN-Net comprehensively fuses the secret and cover images on different scales and depths in the forward hiding process. Besides, the proposed EUIN-Net also maintains the independence of the cover and secret information in the backward revealing process. Moreover, the long-range dependency can be retrieved through using the skip connections between each pair U-shape invertible blocks. The above factors can drive our EUIN-Net to promote the quality of stego and revealed secret images. Furthermore, the shared and multi-scale characteristics of the U-shaped invertible blocks during the hiding and revealing stages contribute to significant reductions of our EUIN-Net in the model size, Flops, and GPU Memory occupancies. Extensive experiments demonstrate that the proposed EUIN-Net can achieve satisfactory performances with limited computational cost for large-capacity image steganography.