A Steganography Method Using Neural Networks

Andres Ali Lopez-Hernandez, Ricardo Francisco Martinez-Gonzalez, Jose Antonio Hernandez-Reyes, Leonardo Palacios-Luengas, Ruben Vazquez-Medina

Research output: Contribution to journalArticlepeer-review


© 2003-2012 IEEE. This work proposes a method of steganography without embedding that uses an artificial neural network (ANN) for subliminally communicating some sensitive information using digital images. The proposed method uses the scaled conjugate gradient (SCG) learning to configure an ANN and replicate a secret message from a cover image. The configuration values are sent to destination alongside the cover image, and the secret message is easily recovered. The method was proved using four ANN architectures changing the number of neurons and by using different cover images. This work presents a steganographic method that allows the exchange of sensitive information between two entities. This method has no capacity limitations, it does not produce perceptibility features of the sensitive information in the cover image and it includes an implicit mechanism of integrity/authentication of the cover image. To estimate the performance of the proposed method, the mean square error (MSE) and the peak signal to noise ratio (PSNR) between the recovered secret image and the original secret image, as well as the computation time in the training stage, for artificial neural network architectures with different number of neurons are calculated. Finally, it is shown that the proposed method has better performance when using digital images with large changes in hue. In these cases, the recovered secret image was equal to the original secret image. This opens the possibility of testing the proposed steganographic method to communicate data and not just digital images.
Original languageAmerican English
Pages (from-to)495-506
Number of pages12
JournalIEEE Latin America Transactions
StatePublished - 1 Mar 2020

Fingerprint Dive into the research topics of 'A Steganography Method Using Neural Networks'. Together they form a unique fingerprint.

Cite this