Robustness Enhancement of Quantization-Based Image Watermarking by Preserving Local Spatial Characteristics

This work proposes that, in quantization-based watermarking, the watermark message can be embedded in the Census directions to improve the robustness performance while resisting the block-wise attacks, e.g. JPEG compression. Experimental results indicate the capability of resisting the block- wise attacks can not rival that of resisting the pixel-wise attacks because the local spatial characteristics might be destroyed by the embedding procedure such that the embedded watermark is vulnerable to block-wise attacks. Since the Census transform of an image remains unchanged when the image is distorted by various manipulations, the proposed approach embeds the watermark message in the Census directions to preserve the local spatial characteristics. Simulation results will show the proposed approach indeed improves the robustness performance in withstanding the JPEG compression. Besides, the proposed approach will modify the Census transform so that the watermark extracting process can have almost the same Census directions as the embedding process. Therefore, the side information could be reduced, which makes the proposed approach more applicable.