Dual-Domain Neural-Wavelet Reversible Image Watermarking Based on High-Capacity Data Hiding with Wavelet-Guided Neural Prediction
Dual-Domain Neural-Wavelet Reversible Image Watermarking Based on High-Capacity Data Hiding with Wavelet-Guided Neural Prediction
Simin Mansouri Boroujeni1 Ayoub Taheri2
1) Department of Statistics University of Payame Noor P.O. Box 19395-4697 Tehran, Iran s_
2) Department of Computer Engineering Faculty of IT University of Payame Noor P.O. Box 19395-4697 Tehran, Iran
IOI Code
:
ACCA-ACAZGK
Publication :
4th International Congress on Science, engineering & New Technologies(secongress.com)
Abstract :
This paper presents Dual-Domain Neural-Wavelet Reversible Watermarking (DDN-RW), a hybrid framework that bridges multiresolution wavelet analysis and deep learning-based spatial prediction. By combining 2-Level Discrete Wavelet Transform (2-Level DWT) decomposition with an Backpropagation Artificial Neural Network (ANN), our approach generates high-precision prediction maps on low-frequency structural approximations (LL2) while isolating residual noise in high-frequency directional sub-bands (LH, HL, HH). Experimental validation demonstrates that DDN-RW significantly outperforms spatial-only ANN methods and frequency-only DWT watermarking, achieving a Peak Signal-to-Noise Ratio (PSNR) exceeding 48 dB at a payload capacity of 0.5 bits per pixel (bpp), alongside complete, bit-exact reversibility.
Keywords : dual domain neural wavelet digital image watermarking high capacity data hiding wavelet guided neural prediction