Scholarlink

<p>This manuscript proposes an optimized hybrid digital watermarking approach using Rotation Dilated Invariant Causal Skill Convolution Attention Networks (RDInv-2CSAN) to enhance watermark security while preserving image quality. The approach solves major problems in watermarking, including invisibility and resistance to attack, including noise injection, compression, and geometric deformations. This is done by first pre-processingggg cover images of CIFAR dataset and MS-COCO dataset using Adaptive Gradient Domain Guided Image Filtering to enhance image quality. The feature extraction and grid line calculation are then done in RDInv-2CSAN and best grid lines are chosen to be used in watermarking. The method uses Dilated Causal Convolution Networks and Rotation-Invariant Attention Networks, which are trained using a Skill Optimization Algorithm (SOA), to improve feature selection. The implementation of watermark embedding applies Adaptive Tunable Q Wavelet Transform and Guaranteed Functional Tensor Singular Value Decomposition, which guarantees strength towards image damage. The performance of the method is proven by experiments with high Structural Similarity Index (SSIM) 99 percent and maximum Signal-to-Noise Ratio (PSNR) 90.45&#xa0;dB which presents high image quality and is undetectable. The proposed method offers significant resistance to various attacks while optimizing feature selection for strong and secure watermarking.</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

Robust Digital Image Watermarking Using Rotation-Dilated Invariant Causal Skill Convolution Attention Networks

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • B. Lalitha,
  • R. Rubesh Selvakumar,
  • S. Priyadharshini,
  • E. Surya

摘要

This manuscript proposes an optimized hybrid digital watermarking approach using Rotation Dilated Invariant Causal Skill Convolution Attention Networks (RDInv-2CSAN) to enhance watermark security while preserving image quality. The approach solves major problems in watermarking, including invisibility and resistance to attack, including noise injection, compression, and geometric deformations. This is done by first pre-processingggg cover images of CIFAR dataset and MS-COCO dataset using Adaptive Gradient Domain Guided Image Filtering to enhance image quality. The feature extraction and grid line calculation are then done in RDInv-2CSAN and best grid lines are chosen to be used in watermarking. The method uses Dilated Causal Convolution Networks and Rotation-Invariant Attention Networks, which are trained using a Skill Optimization Algorithm (SOA), to improve feature selection. The implementation of watermark embedding applies Adaptive Tunable Q Wavelet Transform and Guaranteed Functional Tensor Singular Value Decomposition, which guarantees strength towards image damage. The performance of the method is proven by experiments with high Structural Similarity Index (SSIM) 99 percent and maximum Signal-to-Noise Ratio (PSNR) 90.45 dB which presents high image quality and is undetectable. The proposed method offers significant resistance to various attacks while optimizing feature selection for strong and secure watermarking.