JSN: Design and Analysis of JPEG Steganography Network

Abstract

Image steganography involves hiding a secret message within an image for covert communication, allowing only the intended recipient to extract the hidden message from the “stego” image. The secret message can also be an image itself to enable the transmission of more information, resulting in applications where one image is concealed within another. While existing techniques can embed a secret image of similar size into a cover image with minimal distortion, they often overlook the effects of lossy compression during transmission, such as when saving images in the commonly used JPEG format. This oversight can hinder the extraction of the hidden image. To address the challenges posed by JPEG compression in image steganography, we propose a JPEG Steganography Network (JSN) that leverages a reversible deep neural network as its backbone, integrated with the JPEG encoding process. We utilize 8×8 Discrete Cosine Transform (DCT) and consider the quantization step size specified by JPEG to create a JPEG-compliant stego image. We also discuss various design considerations and conduct extensive testing on JSN to validate its performance and practicality in real-world applications.