A Tree-structured Codebook Partitioning Technique for VQ-based Image Watermarking

摘要

We propose a pairwise codebook partitioning technique for robust watermarking based on vector quantization. We group codewords in pairs by similarity. One is labeled '0' and the other labeled '1'. For each image block, we first find the pair that owns the nearest codeword. The watermark is then embedded by selecting from the pair the codeword whose label equals the corresponding watermark bit. The extraction process can be performed blindly. Simulation results show that the proposed method can survive JPEG compression, high-quality VQ compression and some common signal processing operations.