An End-to-End Video Steganography Network Based on a Coding Unit Mask

Abstract

Steganography hides secret messages inside the covers while ensuring imperceptibility. Different from traditional steganography, deep learning-based steganography has an adaptable and generalized framework without needing expertise regarding the embedding process. However, most steganography algorithms utilize images as covers instead of videos, which are more expressive and more widely spread. To this end, an end-to-end deep learning network for video steganography is proposed in this paper. A multiscale down-sampling feature extraction structure is designed, which consists of three parts including an encoder, a decoder, and a discriminator network. Furthermore, in order to facilitate the learning ability of network, a CU (coding unit) mask built from a VVC (versatile video coding) video is first introduced. In addition, an attention mechanism is used to further promote the visual quality. The experimental results show that the proposed steganography network can achieve a better performance in terms of the perceptual quality of stego videos, decoding the accuracy of hidden messages, and the relatively high embedding capacity compared with the state-of-the-art steganography networks.