Scholarlink

<p>Advancements in multimedia technologies have greatly improved the ease and efficiency of daily activities and professional endeavors, but these advancements have also contributed to an increase in security vulnerabilities in data transmission. In this research, a cutting-edge video steganography framework was introduced, employing adversarial and hierarchical feature learning using post-quantum cryptography to maximize data embedding capacity and enhance security against both classical and quantum computers. Conventional steganographic techniques frequently encounter challenges in balancing the amount of data that could be hidden, the degree to which the hidden data is undetectable, and the ability to withstand attacks. The proposed framework addressed these issues by integrating 3D convolutional neural networks (3D CNNs) for fast feature extraction and probability mass function generative adversarial networks (PMF-GANs) for enhancing the stability and quality of GAN training and outputs. With the use of Nth-degree truncated polynomial ring units (NTRU) in the proposed framework, the challenging lattice problems are sorted that quantum and classical computers find difficult to resolve. The results suggested that the 3D NN-based enhanced video steganography framework trades capacity for image quality and also demonstrated that PMF-GAN produces better visual outcomes than baseline models while attaining improved evaluation performance. Evaluations on the DIVerse 2&#xa0;K resolution (DIV2K) and Common Objects in Context (COCO) datasets show peak-signal-to-noise ratio (PSNR) values of 62.85 and 59.52 and mean squared error (MSE) values of 0.032 and 0.0998, and image fidelity (IF) values of 1.45 and 0.999, respectively, while the structural similarity index (SSIM) values are 0.999 for the DIV2K and 0.998 for the COCO dataset. The suggested framework showed better performance than contemporary PSNR and SSIM-based models during the comparison process. Experiments verify that the proposed framework achieves superior PSNR performance by 15% for DIV2K as well as by 10% for the COCO dataset as compared to the state of the art frameworks. The proposed framework surpassed all previous models regarding SSIM metrics delivering 6% stronger performance on DIV2K and 7% on COCO dataset.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Probability Mass Function-GAN Enhanced Feature Compression for Quantum Resilient and Imperceptible Video Steganography

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Kiran Malik,
  • M. Ravinder,
  • Gaurav Indra

摘要

Advancements in multimedia technologies have greatly improved the ease and efficiency of daily activities and professional endeavors, but these advancements have also contributed to an increase in security vulnerabilities in data transmission. In this research, a cutting-edge video steganography framework was introduced, employing adversarial and hierarchical feature learning using post-quantum cryptography to maximize data embedding capacity and enhance security against both classical and quantum computers. Conventional steganographic techniques frequently encounter challenges in balancing the amount of data that could be hidden, the degree to which the hidden data is undetectable, and the ability to withstand attacks. The proposed framework addressed these issues by integrating 3D convolutional neural networks (3D CNNs) for fast feature extraction and probability mass function generative adversarial networks (PMF-GANs) for enhancing the stability and quality of GAN training and outputs. With the use of Nth-degree truncated polynomial ring units (NTRU) in the proposed framework, the challenging lattice problems are sorted that quantum and classical computers find difficult to resolve. The results suggested that the 3D NN-based enhanced video steganography framework trades capacity for image quality and also demonstrated that PMF-GAN produces better visual outcomes than baseline models while attaining improved evaluation performance. Evaluations on the DIVerse 2 K resolution (DIV2K) and Common Objects in Context (COCO) datasets show peak-signal-to-noise ratio (PSNR) values of 62.85 and 59.52 and mean squared error (MSE) values of 0.032 and 0.0998, and image fidelity (IF) values of 1.45 and 0.999, respectively, while the structural similarity index (SSIM) values are 0.999 for the DIV2K and 0.998 for the COCO dataset. The suggested framework showed better performance than contemporary PSNR and SSIM-based models during the comparison process. Experiments verify that the proposed framework achieves superior PSNR performance by 15% for DIV2K as well as by 10% for the COCO dataset as compared to the state of the art frameworks. The proposed framework surpassed all previous models regarding SSIM metrics delivering 6% stronger performance on DIV2K and 7% on COCO dataset.