Scholarlink

This research presents a fresh way to keep communications secure at a time when people are increasingly worried about data protection and privacy. The ideation behind this is to hide video content within an image while keeping the original image looks unaltered. We suggest a system that uses two autoencoders working together, combining convolutional-LSTM with 3D convolutional layers and adversarial learning techniques to make the embedding and extracting process really solid. The system works by compressing video data efficiently and blending it into a cover image using specially designed convolutional structures with different filters. While the extracted video achieves state-of-the-art fidelity with a PSNR of 76.87 and SSIM of 0.83, which beats existing approaches like 3D GAN and Vid-in-Img-RAN [1] by a wide margin. At the same time, the image with the hidden video still looks great with a PSNR of 42, so you can’t tell anything’s been altered in the image.

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

Deep 3D-Spatio-Temporal Networks for Concealing Video with Images through Adversarial Training

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • M. S. Sharath,
  • S. Anirudh,
  • Shamith Chandra,
  • S. Shreedharan,
  • Surabhi Narayan

摘要

This research presents a fresh way to keep communications secure at a time when people are increasingly worried about data protection and privacy. The ideation behind this is to hide video content within an image while keeping the original image looks unaltered. We suggest a system that uses two autoencoders working together, combining convolutional-LSTM with 3D convolutional layers and adversarial learning techniques to make the embedding and extracting process really solid. The system works by compressing video data efficiently and blending it into a cover image using specially designed convolutional structures with different filters. While the extracted video achieves state-of-the-art fidelity with a PSNR of 76.87 and SSIM of 0.83, which beats existing approaches like 3D GAN and Vid-in-Img-RAN [1] by a wide margin. At the same time, the image with the hidden video still looks great with a PSNR of 42, so you can’t tell anything’s been altered in the image.