Neural networks have been shown effective in deep steganography for hiding a full image in another. However, the reason for its success remains not fully clear. Under the existing cover ($C$) dependent deep hiding (DDH) pipeline, it is challenging to analyze how the secret ($S$) image is encoded since the encoded message cannot be analyzed independently. We propose a novel universal deep hiding (UDH) meta-architecture to disentangle the encoding of $S$ from $C$. We perform extensive analysis and demonstrate that the success of deep steganography can be attributed to a frequency discrepancy between $C$ and the encoded secret image.

Neural networks have been shown effective in deep steganography for hiding a full image in another. However, the reason for its success remains not fully clear.

Neural networks have been shown effective in deep steganography for hiding a full image in another. However, the reason for its success remains not fully clear. Under the existing cover ( C) dependent deep hiding (DDH) pipeline, it is challenging to analyze how the secret ( S) image is encoded since the encoded message cannot be analyzed independently. We propose a novel universal deep hiding (UDH) meta-architecture to disentangle the encoding of S from C . We perform extensive analysis and demonstrate that the success of deep steganography can be attributed to a frequency discrepancy between C and the encoded secret image.

The authors present a new steganography technique based on deep neural networks to simultaneously conduct hiding and revealing as a pair. The main idea is to combine two images of the same size together. The trained process aims to compress the information from the secret image into the least noticeable portions of the cover image and consists of three processes: a prep-Network for encoding features, the Hiding Network creates a container image, and a Reveal Network for decoding the transmitted container image. On the positive side, the proposed technique seems novel and clever, although it uses/modifies existing deep learning frameworks and therefore should be viewed as an application paper. The experiments are comprehensive and the results are convincing.