Reviews: Generating steganographic images via adversarial training

The authors studied how to use adversarial training to learn the encoder, the steganalyzer and the decoder at the same time using unsupervised learning. Specifically, the authors designed an adversarial game of three parties. The encoder generates images based on the cover and the message. Then the generated image can be received to both decoder and steganalyzer. The goal of the encoder and decoder is to correctly encode and decode the message, and the goal of the steganalyzer is to determine whether the image is encrypted.