Scholarlink

<p>To address the challenges of low robustness and high computational complexity in existing medical image privacy protection algorithms, we present a novel approach integrating an enhanced version of the lightweight neural network MobileNetV3 with watermarking technology to propose a zero-watermarking algorithm for medical images, facilitating its deployment on mobile devices. Firstly, our algorithm uses the improved lightweight neural network MobileNetV3 to obtain multiscale deep features of medical images and creates corresponding binary feature vectors using a perceptual hashing algorithm. Then, the watermarking image is encrypted using a Logistic chaotic system to enhance its security. Finally, the encrypted watermarking image is operated on with binary feature vectors to form a zero-watermarking. The experimental results display that our algorithm has high normalized correlation (NC) values under different degrees of attacks, and can effectively resist different types of attacks. In comparison with existing medical image watermarking algorithms, our proposed method exhibits enhanced robustness and reduced computational complexity, effectively safeguarding medical image privacy. This advancement holds considerable potential for the practical implementation of mobile health applications.</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

Zero-Watermarking of Medical Images Based on Improved Lightweight Neural Network MobileNetV3

  • 期刊论文

  • 发表日期：

    2025年3月11日

查看全文
  • Qiuni Li,
  • Wen Zhang,
  • Shangqing Liu,
  • Taocui Yan,
  • Jinglong Du,
  • Baoru Han

摘要

To address the challenges of low robustness and high computational complexity in existing medical image privacy protection algorithms, we present a novel approach integrating an enhanced version of the lightweight neural network MobileNetV3 with watermarking technology to propose a zero-watermarking algorithm for medical images, facilitating its deployment on mobile devices. Firstly, our algorithm uses the improved lightweight neural network MobileNetV3 to obtain multiscale deep features of medical images and creates corresponding binary feature vectors using a perceptual hashing algorithm. Then, the watermarking image is encrypted using a Logistic chaotic system to enhance its security. Finally, the encrypted watermarking image is operated on with binary feature vectors to form a zero-watermarking. The experimental results display that our algorithm has high normalized correlation (NC) values under different degrees of attacks, and can effectively resist different types of attacks. In comparison with existing medical image watermarking algorithms, our proposed method exhibits enhanced robustness and reduced computational complexity, effectively safeguarding medical image privacy. This advancement holds considerable potential for the practical implementation of mobile health applications.