Secure Secret Image Hiding and Protection Using a Two-Channel Deep Hiding Network

The use of digital images has been growing in recent years in various fields of remote sensing, industrial monitoring, social media, defence and other fields where secret information is present, demanding the securing of the secret image. Hence, image steganography offers a solution to embed the image into any arbitrary or random image, but it has some drawbacks in the steganography capacity, quality of recovery in the process of visualisation, and resistance to image processing attacks. In recent years, deep learning in combination with image hiding has attracted a lot of attention. In this review, much attention is focused on the recently proposed deep learning-based techniques including convolutional neural networks, Encoder-Decoder architecture, generative adversarial networks and attention mechanism to enhance the hidden image capacity and the residual unseen of the concealed image quality. This review also concentrates on the so called high capacity image hiding and hiding a secret image. The salient features of the deep learning-based image hiding techniques are summarized and the representative studies like Deep Steganography, HiDDeN, Improved FC-DenseNet, Two-Channel Deep Hiding Network and GAN-based methods are reviewed and comparatively discussed focusing on various aspects of these methods including hidden image capacity, imperceptibility, quality of reconstructed image, and robustness of hidden image. The recent developments on remote sensing image based steganography are briefly reviewed; the attacks like JPEG compression, cropping, resizing and noise towards deep learning based steganography are also discussed. Deep neural networks have been observed to enhance the image capacity better than a few traditional approaches, and they can enhance the image reconstruction quality, as well. However, there are still a lot of challenges in the field of deep learning image hiding. Therefore, attention guided adversarial mechanisms and multi-stage fusion are anticipated to be a potentially fruitful area for future research.