Scholarlink

<p>The protection of image integrity and authenticity has become critical due to the growing use of digital data in numerous situations. This work presents a novel method for enhancing resilience and security in digital image watermarking by using deep learning algorithms. In this research, a zero watermarking strategy is presented with an enhanced GhostMob-Net model to overcome the drawbacks of the current image watermarking techniques. This model proposes embedding the hidden message into the cover image and retrieving it from the original image. Then, the Lifting wavelet transform (LWT) is used to extract the low-frequency data from the watermarked image. After that, a two-dimensional (2D) Henon chaos mapping procedure is used to encrypt the watermarked images. In addition, the proposed method uses the hybrid optimization algorithm Gannet Aquila (GaAq) to reduce information loss during the training phase and increase the classifier’s generalization capabilities. By combining both techniques, a multi-layered defense mechanism and resistance boosting to frequent attacks can be provided with the GaAq optimized GhostMob_Net model. This research is evaluated using different testing parameters, which demonstrate its robustness against various image attacks. For image quality analysis, assessment measures include Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Mean Squared Error (MSE). In contrast, security analysis uses the Number of Pixel Change Rates (NPCR) and Unified Average Changing Intensity (UACI). The performance of the proposed method is compared to that of existing digital watermarking techniques, demonstrating its superior capacity to secure image data. This study conducts its research using the USC-SIPI dataset. The proposed model is compared with several existing models in terms of SSIM, MSE, and PSNR. At the embedding stage, the proposed model obtained an average PSNR of 31.14, an average SSIM of 0.949346, and an MSE of 50.42. At the retrieval stage, the proposed model had an average PSNR of 36.14, an average SSIM of 0.957655, and an MSE of 40.4. In addition, the proposed methodology achieves a lower Normalized Correlation (NC) value of 0.759 in 10 Gaussian attacks, 0.885 in 25 Median attacks, and 0.94 in 10 rotation angle attacks than previous methods. These results demonstrate that the proposed model clearly outperforms all other existing models in terms of both zero watermarking strategies as well as security and significantly enhances digital image watermarking quality and robustness.</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

Boosting security and resilience in digital color image zero watermarking with the Gannet Aquila-Inspired ghost mob_net technique

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • Sambhaji Marutirao Shedole,
  • Santhi Vaithiyanathan

摘要

The protection of image integrity and authenticity has become critical due to the growing use of digital data in numerous situations. This work presents a novel method for enhancing resilience and security in digital image watermarking by using deep learning algorithms. In this research, a zero watermarking strategy is presented with an enhanced GhostMob-Net model to overcome the drawbacks of the current image watermarking techniques. This model proposes embedding the hidden message into the cover image and retrieving it from the original image. Then, the Lifting wavelet transform (LWT) is used to extract the low-frequency data from the watermarked image. After that, a two-dimensional (2D) Henon chaos mapping procedure is used to encrypt the watermarked images. In addition, the proposed method uses the hybrid optimization algorithm Gannet Aquila (GaAq) to reduce information loss during the training phase and increase the classifier’s generalization capabilities. By combining both techniques, a multi-layered defense mechanism and resistance boosting to frequent attacks can be provided with the GaAq optimized GhostMob_Net model. This research is evaluated using different testing parameters, which demonstrate its robustness against various image attacks. For image quality analysis, assessment measures include Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Mean Squared Error (MSE). In contrast, security analysis uses the Number of Pixel Change Rates (NPCR) and Unified Average Changing Intensity (UACI). The performance of the proposed method is compared to that of existing digital watermarking techniques, demonstrating its superior capacity to secure image data. This study conducts its research using the USC-SIPI dataset. The proposed model is compared with several existing models in terms of SSIM, MSE, and PSNR. At the embedding stage, the proposed model obtained an average PSNR of 31.14, an average SSIM of 0.949346, and an MSE of 50.42. At the retrieval stage, the proposed model had an average PSNR of 36.14, an average SSIM of 0.957655, and an MSE of 40.4. In addition, the proposed methodology achieves a lower Normalized Correlation (NC) value of 0.759 in 10 Gaussian attacks, 0.885 in 25 Median attacks, and 0.94 in 10 rotation angle attacks than previous methods. These results demonstrate that the proposed model clearly outperforms all other existing models in terms of both zero watermarking strategies as well as security and significantly enhances digital image watermarking quality and robustness.