Convolutional neural network and wavelet composite against geometric attacks a watermarking approach.

Area of Science:

• Digital watermarking
• Information security
• Image processing

Background:

• E-governance systems require robust ownership identification for services like banking, healthcare, and insurance.
• Traditional watermarking methods face challenges in balancing payload capacity, robustness, and perceptual transparency.
• Authentication is critical for secure data exchange in digital environments.

Purpose of the Study:

• To propose a novel three-layer, feature-dependent image watermarking scheme in the transform domain.
• To enhance security and authentication in digital e-governance applications.
• To achieve high payload capacity, perceptual transparency, and robustness against various attacks.

Main Methods:

• A three-layer watermarking approach using Discrete Wavelet Transform (DWT) was developed.
• Singular Values of an encrypted logo, arithmetic coding for textual signatures, and run-length coding for owner identity were embedded in different DWT decomposition levels.
• Feature extraction and data compression techniques were utilized for embedding.
• Convolutional Neural Network (CNN) was employed for cover image reversibility and watermark extraction.
• Attacks including Gaussian noise, salt and pepper noise, rotation, and cropping were simulated.

Main Results:

• The proposed scheme achieved high perceptual transparency, with Structural Similarity Index Measure (SSIM) and Normalised Correlation (NC) approaching unity.
• Robustness against various attacks was demonstrated, with a Peak Signal-to-Noise Ratio (PSNR) of approximately 44 dB.
• High SSIM (>0.99) and NC (~1.0) values confirmed enhanced perceptual transparency and robustness.
• Minimal bit error rates post-attack indicated reliable watermark recovery with CNN-aided restoration.
• Low Mean Square Error (MSE) was achieved during cover image recovery.

Conclusions:

• The developed DWT-based watermarking scheme effectively embeds a heavy payload while maintaining high perceptual transparency and robustness.
• The integration of CNN significantly enhances the scheme's resilience against diverse attacks and aids in accurate image restoration.
• This approach offers a promising solution for secure authentication in e-governance and other sensitive digital applications.