Abstract:  A new blind watermarking scheme for color image was proposed. In this scheme, a binary logo watermark was embedded into the discrete cosine transform (DCT) domain of the color image with the help of radial basis function (RBF) neural networks. According to HVS, the original image was transformed from the RGB color space to the brightness and chroma (YCrCb) space. In order to strengthen the imperceptibility and robustness, the watermark was embedded into the DC coefficient of 8×8 blocks in CrCb chroma components. A key was set to select blocks where the watermark was embedded, which can improve the security of the watermark. Also, RBF was used to learn and approximate the inverse procedure of quantization, and watermark embedding and extraction was performed with the trained neural network to further improve the robustness. Simulations show that the proposed method has good imperceptibility as well as robustness to common image processing attacks.

Key words: digital watermark, color image watermark, RBF neural network, YCrCb color space