Scholarlink

<p>Image watermarking is a dynamic field of research focused on preventing unauthorized use and alteration of images. In reversible watermarking, it’s essential to consider both the robustness of the watermark and the privacy preservation of the cover image. However, Joint Photographic Expert Group (JPEG) compression and noise addition are among the few robust reversible watermarking techniques that could resist common attacks while simultaneously preserving privacy. To tackle these issues, an innovative technique named Sand Cat Hunter Optimization_ LeNet+ Haar Wavelet Transform (SCHO_LeNet+HWT) is developed for reversible image watermarking. Here, SCHO is the combination of Sand Cat Swarm Optimization (SCSO) and Hunter–Prey Optimizer (HPO). This model comprises two phases: the embedding phase and the extraction phase. In the embedding phase, the gridding process is done by receiving the input image, which is followed by Haar Wavelet Transform (HWT). Then, the selection of the interested region is done using LeNet, which is trained by SCHO. After that, the selected interested region and the watermark image are embedded in the embedding phase. Then, inverse HWT is applied and merged to form an embedded watermark image. At the extraction phase, the embedded watermark image is considered as the input, which is followed by HWT. Then, the extraction process is conducted along with the reversible image process. Afterwards, the keymap (interested region) is given to both the reversible image process and the extracted watermark image for recovering the original image. The effectiveness of SCHO_LeNet+HWT is examined by considering parameters, such as Peak Signal-To-Noise Ratio (PSNR), Normalized&#xa0;Correlation&#xa0;(NC), and Bit&#xa0;Error&#xa0;Rate&#xa0;(BER). The investigations revealed that the proposed SCHO_LeNet+HWT offered high robustness against various noises and attacks and computed a maximum PSNR, NC, of 35.519dB, and 0.882, respectively, and obtained the least BER value of 0.042.</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

Optimized deep learning based interested region selection for reversible image watermarking using Haar wavelet transform

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Senthilraja B,
  • Anandan P,
  • Anbukaruppusamy S

摘要

Image watermarking is a dynamic field of research focused on preventing unauthorized use and alteration of images. In reversible watermarking, it’s essential to consider both the robustness of the watermark and the privacy preservation of the cover image. However, Joint Photographic Expert Group (JPEG) compression and noise addition are among the few robust reversible watermarking techniques that could resist common attacks while simultaneously preserving privacy. To tackle these issues, an innovative technique named Sand Cat Hunter Optimization_ LeNet+ Haar Wavelet Transform (SCHO_LeNet+HWT) is developed for reversible image watermarking. Here, SCHO is the combination of Sand Cat Swarm Optimization (SCSO) and Hunter–Prey Optimizer (HPO). This model comprises two phases: the embedding phase and the extraction phase. In the embedding phase, the gridding process is done by receiving the input image, which is followed by Haar Wavelet Transform (HWT). Then, the selection of the interested region is done using LeNet, which is trained by SCHO. After that, the selected interested region and the watermark image are embedded in the embedding phase. Then, inverse HWT is applied and merged to form an embedded watermark image. At the extraction phase, the embedded watermark image is considered as the input, which is followed by HWT. Then, the extraction process is conducted along with the reversible image process. Afterwards, the keymap (interested region) is given to both the reversible image process and the extracted watermark image for recovering the original image. The effectiveness of SCHO_LeNet+HWT is examined by considering parameters, such as Peak Signal-To-Noise Ratio (PSNR), Normalized Correlation (NC), and Bit Error Rate (BER). The investigations revealed that the proposed SCHO_LeNet+HWT offered high robustness against various noises and attacks and computed a maximum PSNR, NC, of 35.519dB, and 0.882, respectively, and obtained the least BER value of 0.042.