Novel robust multiple watermarking against regional attacks of digital images

Abstract

This paper presents a novel robust multiple watermarking method for regional attacks of digital images. The core idea of the proposed method is to divide the host image into 4?×?4 non-overlapping regions and embed the watermark image into these regions repeatedly. First, the binary watermark image is scrambled by Arnold transform and divided into four equal parts. Then, the variance and mean value information of each 8?×?8 size block in the host image is used as side information, and each part of the scrambled watermark is embedded in the particular four regions of the host image. In the watermark extraction process, each part of the scrambled watermark image is divided into four equal sub-parts further. The similarity between the sub-part of the scrambled watermark and each extracted watermark segment corresponding to this sub-part is used as side information to search the final extracted watermark segment. For ensuring the visual quality of the watermarked image, the watermark embedding quantization steps are selected by combining Human Visual System (HVS) and Particle Swarm Optimization (PSO). Moreover, a novel watermark quality evaluating measurement, called Weighted Normalized Correlation (WNC), is proposed. Experimental results demonstrate good visual imperceptibility and robustness of the proposed method against traditional regional attacks, accidental attacks and joint attacks, which are performed by Stirmark or Adobe Photoshop CS5.