Reversible video steganography based on residual structure and attention mechanism

In video steganography, a hidden video is embedded into a carrier video and then extracted from the carrier video after a covert transmission. Using a mixed attention mechanism and residual structure, this research proposes a reversible video steganography network that greatly enhances the quality of steganography videos and secret video retrieval. Using a reversible neural network as its foundation, the network uses a dual-channel structure, extracts deep features using a mixed attention mechanism, and encourages information interaction between various features via channel exchange. This study introduces residual structure to improve the accuracy of information recovery by allowing features to be reused and introducing comparable features into the features lost during training. According to the testing findings, the framework outperforms the comparison algorithm in several evaluation indices and is capable of securely concealing and extracting secret video without causing loss.