A new DCT-based digital image watermarking algorithm

In this paper, we propose a novel digital image watermarking algorithm, which can embed a multi-bit message into the image and realize the blind extraction of the message. To embed the watermark, the image is firstly scrambled to enhance the security. Then, some feature vectors based on the DCT of the scrambled image are extracted. The expanded watermark generated using the spread spectrum technique is embedded into the extracted feature vectors. The Watson perceptual model is also considered to obtain good invisibility. By means of corresponding inverse transformations we get the watermarked image. The watermark extraction process is similar as the embedding process. Unlike previous works, we use a new scheme to extract the image feature vectors and embed the watermark repeatedly into them. And the codebook in spread spectrum watermark generation is orthogonalized and unitized. These steps can dramatically improve the robustness of the watermark. Experiment results demonstrate the robustness of our algorithm against various common attacks such as JPEG compression, additive noise, histogram equalization, low-pass filtering and cropping.