A Virtual Primary Key Generation Scheme Based on Dimensionality Reduction and Encoding for Digital Fingerprinting

The relational database fingerprinting technique using virtual primary keys guarantees the fingerprint embedding process of tuples without primary keys. However, the virtual primary key generated by the existing scheme has the problems of high redundancy and easy destruction, which seriously affects the effect of fingerprint recognition. This paper proposes a new scheme based on variational auto-encoder (VAE) reduction and median tree coding. The dimensionality reduction process reduces the dependence of the scheme on the value of a single attribute, thus improving the robustness. Median tree coding ensures a uniform distribution of generated virtual primary keys to reduce redundancy. Therefore, the stability of fingerprint embedding and the success rate of fingerprint recognition are improved. Experiments are carried out on data sets with 10 and 98 attributes, respectively. It has been proved that the performance of our scheme has greatly improved compared with the existing schemes. Our scheme also shows resistance to various attacks such as attribute deletion, tuple deletion, and sorting attacks.