A Mellin transform based video steganography with improved resistance to deep learning steganalysis for next generation networks

Abstract

In the era of 5 G network advancements, the potential for extremely robust, less-latency, and huge-capacity communication opens up new perspective for multimedia. Steganography enables embedding of sensitive data within multimedia files, making it unreadable to unauthorized third parties. Notably, when using videos as cover, the capacity for data embedding is substantially increased. Recent developments in steganography have largely revolved around modified versions of transform domain techniques. Due to this repetitiveness, it becomes easier for steganalytic tools in detecting concealed data. Addressing this issue, our paper introduces an innovative data embedding approach MARVIS based on the Mellin transform. The superiority of the proposed approach is exhibited using the metrics, MSE, PSNR, and SSIM. MARVIS has achieved PSNR of 50–60 dB and SSIM of 0.9998 for embedding 4 bits of secret data, outperforming other methods that achieve 40 dB for 1 bit. By quadrupling stego capacity, we can embed more secret data per pixel without compromising the integrity of the cover object.

• •

  MARVIS utilizes phase modulation for data embedding, offering advantages beyond traditional frequency domain techniques which use frequency domain for data embedding.

• •

  The effectiveness of the proposed data embedding approach is validated through Y-Net, a deep learning-based steganalysis tool.