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

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 embedding process. However, most algorithms utilize images as instead of videos, which are more expressive widely spread. To this end, end-to-end learning network for video is proposed in paper. A multiscale down-sampling feature extraction structure designed, consists three parts including encoder, a decoder, discriminator network. Furthermore, order to facilitate ability network, CU (coding unit) mask built VVC (versatile coding) first introduced. In addition, attention mechanism used further promote visual quality. The experimental results show that can achieve better performance terms perceptual quality stego decoding accuracy hidden messages, relatively high capacity compared with state-of-the-art networks.