This paper proposes StegOT, a steganographic model that hides a secret image in a cover image of the same resolution. StegOT is an autoencoder-based model that leverages optimal transport theory to address the mode collapse problem of generative adversarial networks (GANs) and variational autoencoders (VAEs). A multi-channel optimal transport (MCOT) module is designed to transform multi-peak feature distributions into a single peak, thereby achieving information balance. Experiments demonstrate that StegOT achieves a balance between the cover and secret images, and improves the quality of both the hidden and restored images.
