Image Steganography: Uncovering Hiding Image Using Deep Learning for Attention-Generative Adversarial Network

Steganography is one of the most crucial role in the field of information hiding, which is the technique to hiding a secrete image from ordinary file or message in order to avoiding the detection of steganography models and human eye. In recent years, every scholars have various deep learning networks to the fields of steganography to improving accuracy of detection. In this paper propose Attention- Generative Adversarial Network (GAN) has two aspects: Architecture of Network and Loss function. Architecture network consists of three networks: encoder network, decoder network and discriminator network. Encoder has a hiding the secrete color image into a cover color image with the same size to generate the steganography image from a visual quality by encoder network. Secrete image are extracted by the decoder network. Discriminator taking the cover image and stego image as input to classify the whether the input image contain the secrete image. Finally, Loss function, they are optimizing the weight of the encoder and decoder network through an adversarial training model. The results obtained DL using Attention-GAN show a high PSNR was improved to 44.60, SSIM of 0.998, and MSE of 103.26 on ImageNet dataset, which ensure GAN compared to other exiting methods like Light-weight method, Deep Adaptive Hiding Network (DAH-Net) and Transformer Recursive Permutation (TRP) stego method.