A hierarchical protection scheme for intellectual property of semi-open source datasets based on double watermarking

The high quality datasets are an important guarantee for the rapid development of deep learning, so the protection of datasets should be taken seriously. The semi-open source datasets are the datasets that are free for some users and used for academic communication and are charged for another part of users and generates commercial value. However, most of the existing datasets protection mechanisms are only a single passive copyright verification mechanism, which cannot protect semi-open source datasets well. Based on this situation, this paper proposes a hierarchical protection scheme for the intellectual property of semi-open source datasets based on double watermarking. The scheme adopts different protection schemes for datasets for the above mentioned user groups: internal users and purchased users, to achieve the purpose of hierarchical protection. The passive verification and active protection functions for semi-open source datasets are realized by embedding blind watermark and visual watermark, respectively. Finally, the paper conducts experiments on a number of underlying datasets to verify the imperceptibility, universality, harmlessness and effectiveness of the scheme.