Deep Secrecy: Advancing Image Steganography Via Deep Learning Technique

Abstract

In today's internet-driven world, while seamless connectivity and social media make tasks easier, they also bring risks like hacking and data theft. Steganography is an emerging field that hides various data types such as images, videos, texts, and audio to protect sensitive information shared online. This term, derived from the Greek words "steganos" (hidden) and "graph" (writing/drawing), reflects the essence of this practice: hiding messages in plain sight. This technique combines science and art to ensure information concealment, addressing the growing need for secure data exchange. To deal with the data theft issues the proposed methodology presents a novel Image Steganography technique using 2 networks one is the hiding network and the other is the revealing network built on the foundation of CNN-based UNeT Architecture. Equipped with specialized encoder and decoder components, the proposed model masterfully hides secret pictures within cover images, all while preserving the cover image's natural quality. The hiding network hides the secret image within the cover image and generates the stegano image while the revealing network extracts the secret image from the cover image and reveals the original secret image. The dataset used in this research is the “COCO dataset” from where random 7500 images were used to train the model. The proposed model successfully produced a Peak Signal Noise Ratio (PSNR) value of 41.08 db. The goal is to produce undetectable stego images making its detection practically difficult.