Image Steganography Using Deep Learning Methods to Detect Covert Communication in Untrusted Channels

Abstract

Media has become a primary medium of communication with the help of the constantly evolving technology. Social media like Facebook, YouTube, Twitter, WhatsApp, and other sites have become a platform for exchanging audio, video, and text messages. This has left them vulnerable to attacks which makes it essential to protect the confidential messages sent over the media channel. Image steganography is the procedure used for camouflaging a secret image in a cover image. In contrast, the method of detecting and extracting the secret information from the stego image is called steganalysis. Steganography can be used positively to secure the data transmission process. On the other hand, it can be used adversely by hackers, criminals, and covert operators for the secret exchange of messages. This work aims to develop a steganography model to embed the secret image and steganalysis tool to extract the embedded secret image. In recent times, deep learning methods have gained popularity and are widely used in the field of steganography. In this work, a unique auto encoder-decoder with a deep convolutional neural network is proposed. Training and testing are done on a subset of the COCO, CelebA, and ImageNet dataset. To evaluate the proposed method, Peak Signal-to-Noise Ratio (PSNR) and Mean Squared Error (MSE) metrics are used. The proposed method has proved to achieve higher invisibility, security, and robustness.