IMPROVED COVERLESS INFORMATION HIDING UTILIZING GENERATIVE MODELS

The submitted algorithm increased data-hiding capacity, extended capability, and improved quality of the reconstructed secret image as compared with traditional steganography and existing coverless information-hiding methods. Unlike conventional steganography, in which the secret information is embedded into a cover image, our method involves the design of generative models that generate a cover image independently of the secret image; thus, there is no cover image explicitly defined. Two generative models are thus used in the proposed method: F creates a middle image with respect to the secret image, while G reconstructs the secret image from the generated cover image. The content consistency extraction module thus encodes content information from the secret image into the generated cover image so as to address the issues of color distortion and content loss in the reconstructed secret image. The experimental results confirm that our method embeds more data, exhibits larger security, and has higher visual quality based on improved PSNR (30.2313), SSIM (0.7762), and reduced MSE(40.1271) values. This method used a dataset comprising 21,550 images, with its results showing vast improvement compared to existing methods