[Retracted] Deep Image Watermarking to JPEG Compression Based on Mixed-Frequency Channel Attention

Figure 1

Overall model architecture. The message processor can learn the method for expanding message and realizing redundancy by the transposed convolutional layer; the encoder includes the mixed-frequency channel attention block which embeds the secret message into the whole cover image, the noise layer changes the kind of noise according to the MBRS method for offering the robustness, and the decoder extracts the secret message from the encoded image. An adversary discriminator is used to distinguish the cover image and the encoded image.