ENDGAN: An End-to-End Robust Video Steganography Using Encoder-Decoder Cross Association Graph-Based Generative Adversarial Network

Video steganography is vital in secure communication through hiding supplementary information in the video frames. However, current solutions meet some limitations associated with their reliability, invisibility, and computational complexity for processing dynamic, compressed, and/or transformed videos. These limitations decrease the reliability of message text hidden and affect the level of security. To overcome these challenges, an Encoder-Decoder cross-association graph-based Generative Adversarial Network (ENDGAN) is developed to use graph-based models and cross-association mechanisms to achieve reliable message embedding, extraction, and quality video synthesis. Finally, the Cat Hunting Optimization (CHO) algorithm is introduced to optimize the loss functions of ENDGAN for reliability and speed purposes. This approach improves the accuracy of message recovery, does not affect imperceptibility, and decreases computational time. The experimental analysis reveals that it outperforms the existing techniques for providing safe and effective video steganography inverse transforms and different video distortions. The proposed approach attains a higher accuracy of 99.9%.