In this work, we demonstrate the application of Variational Autoencoder-Generative Adversarial Network (VAE-GAN) in multimedia data hiding. Unlike traditional deep learning techniques using single architectures, this method exploits the advantages of both VAE and GAN in data hiding to devise an integrated and robust information hiding method. We present an end-to-end trainable model of VAE-GAN which can learn to embed a message image onto a cover image. The devised architecture consists of a VAE with adversarial training as embedder and a fully convolutional network with adversarial training as the extractor. The encoded image is subjected to multiple attacks and it is established that the proposed method is robust towards attacks like Gaussian blurring, rotation, noise, and cropping. The one remarkable feature of our method is that it can be trained to recover against various attacks. Hence it is possible to make it more robust by training the network on other popular attacks.
rohitgandikota/Hiding-Images-using-VAE-Genarative-Adversarial-Networks
Variational Autoencoder-Generative Adversarial Network (VAE-GAN) to hide data inside images
Python