Feature-based Image Watermarking Against To Geometrical Attacks

Along with the rapid development of the network technique andmultimedia technique, the storage, duplication and dissemination of thedigital multimedia information become more and more convenient. As aresult, the piracy problem directly threatens the legal interest of the digitalmedia’s originator and owner. Copyright protection of digital media is anactual problem which needs to be addressed. Digital watermarkingtechnology is considered as a good solution for the afore-mentionedproblem and has been a rising research field in the recent years.The study of robust watermarking has made a certain progress in therecent years, but most of the existing watermarking schemes are unable toresist geometric attacks, which is common in image processing. Theresearch of robust image watermarking resistant to geometric attack is thekey to the usage of watermarking technique. Aiming at improving therobustness of image watermarking, this paper makes a deep investigationinto the previous robust image watermarking, and proposes two image watermarking methods based on image feature.1． Robust image watermarking against to geometric attacks basedon image restoration using SIFT. The matched points determined by thescale-invariance feature transform are used for image restoration to reducesynchronizations errors caused by geometric distortions. Watermarkembedding is achieved by modifying the magnitude of a couple of DFTcoefficients. The watermark detection decision is based on the number ofmatched bits between the retrieved and original watermark. Experimentalresults show that the proposed scheme is robust to common signalprocessing and geometric attacks, including rotation, scaling, cropping,shearing and some combined attacks.2． Image watermarking against to geometric attacks based onfeature region. Watermark synchronization is achieved by local featureregion, which determined by SIFT feature, including position, scale andorient ion. In each circle region, watermark is embedded repeatedly in thespatial domain using odds and even quantification. Just-noticeabledifference is also used to control the modified quantity of each pixel. Simulation results show that the proposed scheme is robust to bothcommon signal processing and geometric distortions.This study will not only provide new method and key technique forrobust image watermarking, but also offer methods for watermarking ofother digital media.