Robust hybrid image watermarking scheme based on KAZE features and IWT-SVD

Digital watermarking is a critical technique for digital rights management (DRM). In the last decade, the local feature-based watermarking schemes have attracted much attention because of their remarkable robustness against geometrical attacks, which will destroy the synchronization between watermark embedding and detection. However, they only used local feature and thus their robustness is based on invariances of the particular local features, which still have limitations when suffering various signal processing attacks. To address this problem, a robust hybrid color image watermarking scheme (RHIW) is proposed in our study by fusing a local feature-based watermarking procedure with a traditional transform domain-based watermarking procedure. In the local feature-based procedure, watermarks are embedded into significant bit-planes of the KAZE feature regions repeatedly by modifying their histograms. In the transform domain-based procedure, watermark is embedded into the IWT-SVD domain by modifying the entries of left singular vector matrices. For a queried image, all the watermarks of KAZE-based and IWT-SVD-based procedures are extracted and compared with the original watermark. Then, the comparison results are fused following an attention-based fusion method to identify the copyright ownership. The experimental results demonstrate that RHIW is sufficient robust not only against geometrical attacks such as cropping, row & column removal and rotation attacks, but also against various common signal processing attacks such as filtering, noise addition, brightness modification and contrast modification, which outperforms state-of-the-art local feature-based watermarking schemes.