Scholarlink

<p>This paper proposes a spread spectrum (SS) image watermarking to protect both the object and the background of variable sizes using a convolutional auto-encoder (CAE) for time-critical authentication in defense surveillance on edge-IoT network. CAE represents the latent feature space (attributes) of the scene (cover) image with variable sizes that make the watermark-embedding space (size) compatible to resource-constrained edge nodes. The proposed work also suggests a deep convolutional neural network (DCNN) based classifier to detect whether images are watermarked or not, using spatial patterns as well as statistical features like Kullback–Leibler divergence (KLD) and edge entropy. A large set of simulation results show that peak-signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) values of the watermarked image are &#xa0;45.24 dB and &#xa0;0.9967 for 64 and 128 folds reduction on the object and the background latent space of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\((128 \times 128)\)</EquationSource> </InlineEquation> image, while DCNN offers an accuracy on the watermark detection of <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\sim 0.94\)</EquationSource> </InlineEquation>.</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

CAE-DCNN Architectures for Image Watermarking and Detection on Edge-IoT Networks

  • 期刊论文

  • 发表日期：

    2026年2月24日

查看全文
  • Prasenjit Kumar Patra,
  • Santi P. Maity

摘要

This paper proposes a spread spectrum (SS) image watermarking to protect both the object and the background of variable sizes using a convolutional auto-encoder (CAE) for time-critical authentication in defense surveillance on edge-IoT network. CAE represents the latent feature space (attributes) of the scene (cover) image with variable sizes that make the watermark-embedding space (size) compatible to resource-constrained edge nodes. The proposed work also suggests a deep convolutional neural network (DCNN) based classifier to detect whether images are watermarked or not, using spatial patterns as well as statistical features like Kullback–Leibler divergence (KLD) and edge entropy. A large set of simulation results show that peak-signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) values of the watermarked image are  45.24 dB and  0.9967 for 64 and 128 folds reduction on the object and the background latent space of \((128 \times 128)\) image, while DCNN offers an accuracy on the watermark detection of \(\sim 0.94\) .