Adaptive watermark scheme with RBF neural networks

This paper proposes an adaptive digital watermarking scheme with RBF neural networks, in which a visually recognizable binary image watermark is embedded into the DCT domain of the cover image. The watermark was encrypted by chaotic series and inserted into the middle frequency coefficients of the cover image's blocked DCT-based transform domain. In order to make the watermark stronger to resist different types of attacks, it is important to adapt the embedding maximum amount of interested watermark before the watermark becomes visible. In this paper, RBF neural networks are used to achieve maximum-strength watermark according to the frequency component feature of the cover image. Experimental results show that the proposed techniques have good imperceptibility and can survive of common image processing operations and JPEG lossy compression with high robustness.