A batch copyright scheme for digital image based on deep neural network.

Digital signature and watermarking are effective image copyright protection techniques. However, these methods come with some inherent drawbacks, including the incapacity of carrying information and inevitable fidelity loss, respectively. To improve this situation, this paper proposes a neural network-based image batch copyright protection scheme, with which a copyright message bitstream can be extracted from each registered image while no modifications are introduced. Taking advantage of the pattern extraction capability and the error tolerance of the neural network, the proposed scheme achieves perfect imperceptibility and superior robustness. Moreover, the network's preference for diverse data content makes it especially appropriate for multiple images copyright verification. These claims will be further supported by the experimental results in this paper.

Locating secret messages based on quantitative steganalysis.

Steganography poses a serious challenge to forensics because investigators cannot identify even traces of secret messages embedded using a steganographer. Contrarily, the objective of locating steganalysis is to locate the embedded message, which should help extract the secret message. In this paper, a methodology of locating steganalysis using quantitative steganalysis is presented for multiple stego images with embedded messages along the same embedding path. Three typical quantitative steganalysis methods are applied to the methodology to locate the messages embedded using LSB re-placement. Experimental results show that the presented methods can reliably estimate the embedding positions, which verifies the validity of the presented methodology. The presented methodology points out a new use of quantitative steganalysis, and further demonstrates that it is necessary to design more precise quantitative steganalysis methods.