Scholarlink

<p>Data security is of utmost importance in the medical field. Medical data are prone to security attacks often. The research proposes and executes a deep learning model for digitally watermarking medical images (Chest X ray images) to increase security. The work is done in three phases. In the first phase, the medical image is visibly watermarked with the image using region-of-interest (ROI), invisibly using discrete wavelet transforms (DWT), and with a text using a logistic mapping approach. It involves embedding text and image watermarks in medical images using a deep-learning model. The second phase involves applying noise attacks to images and evaluating their performance by estimating peak signal-to-noise ratio (PSNR) values. In the third phase, the model is used to extract the watermarks. Optimization of the model is done using the Adam optimizer (AOA) and stochastic gradient descent (SGD) algorithms. The loss is estimated using the categorical cross-entropy loss function. The DNN model developed is found to render 94% accuracy in image watermark classification and 97.5% in text identification, respectively. The model’s accuracy obtained in watermark identification is 91.43% with loss value of 0.2. The PSNR value obtained is 39&#xa0;dB-46&#xa0;dB, which is significant and efficient.</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

Enhancing the protection of medical images using watermarking technology processed with deep learning algorithms

  • 期刊论文

  • 发表日期：

    2025年1月18日

查看全文
  • Abirami R,
  • Malathy C

摘要

Data security is of utmost importance in the medical field. Medical data are prone to security attacks often. The research proposes and executes a deep learning model for digitally watermarking medical images (Chest X ray images) to increase security. The work is done in three phases. In the first phase, the medical image is visibly watermarked with the image using region-of-interest (ROI), invisibly using discrete wavelet transforms (DWT), and with a text using a logistic mapping approach. It involves embedding text and image watermarks in medical images using a deep-learning model. The second phase involves applying noise attacks to images and evaluating their performance by estimating peak signal-to-noise ratio (PSNR) values. In the third phase, the model is used to extract the watermarks. Optimization of the model is done using the Adam optimizer (AOA) and stochastic gradient descent (SGD) algorithms. The loss is estimated using the categorical cross-entropy loss function. The DNN model developed is found to render 94% accuracy in image watermark classification and 97.5% in text identification, respectively. The model’s accuracy obtained in watermark identification is 91.43% with loss value of 0.2. The PSNR value obtained is 39 dB-46 dB, which is significant and efficient.