Enhancing image steganography via adversarial optimization of the stego distribution

Hongyue Zha, Weiming Zhang, Nenghai Yu, Zexin Fan

Abstract: Highlights•Stego distribution optimization achieves efficient adversarial steganography.•Optimizing embedding modification probabilities enables end-to-end gradient descent.•Cover enhancement and distortion optimization can be collaborated.•Pairing cover-stego in loss removes mode collapses in steganographic min-max games.