A Deep Belief Network-Based Robust Image Watermarking Method Using Modified Dual-Tree Complex Wavelet Transform and OTP Encryption

作者：Mangalagowri Gorbal, Ramesh Shelke, Manuj Joshi, Dilendra Hiran · 发表于：International Journal of Image and Graphics · 年份：2025 · DOI：10.1142/s0219467827500793

Digital image watermarking, which covertly embeds information within photos while maintaining their visual quality, is essential to protecting intellectual property. This study introduces a new method for image watermarking that makes use of enhanced One-Time Pad (OTP) encryption and the Modified Dual-Tree Complex Wavelet Transform (DT-CWT). The technique guarantees safe watermark extraction and insertion in cover photos while remaining resilient to possible intrusions. The cover image is divided into four parts during the embedding phase. Singular Value Decomposition (SVD) and Modified DT-CWT are then used for effective frequency domain embedding, and improved OTP encryption is used for increased security. Additionally, Modified Arnold Scrambling’s key generation is aided by a Deep Belief Network (DBN), which increases security. Inverse processes of Modified Arnold Scrambling and SVD are used to recover the watermark intact during extraction, whereas inverse processes of Modified DT-CWT, SVD, and OTP decryption are used to recover the original cover image. The DBN approach for Dataset[Formula: see text]1 produces an SSIM of 0.812, which shows that the image’s structure is strongly preserved even after compression. With an SSIM of 0.758 in Dataset[Formula: see text]2, the DBN approach once more outperforms traditional methods such as DNN (0.705), GRU (0.670), and LSTM (0.669).