Secure Deduplication Method Based on Autoencoder

Message lock encryption is the basis of current secure deduplication research. However, since the theoretical basis of randomized Message-locked encryption (R-MLE) is bilinear mapping, the computational overhead is large. Although this method has semantic security and supports data deduplication, it is not practical. This paper solves the core problem that the deduplication method based on R-MLE is not efficient. The method of this paper does not take the improvement of encryption algorithm efficiency as a breakthrough, but introduces the autoencoder commonly used in image processing, which greatly improves the deduplication efficiency by greatly reducing the number of tag comparisons. Theoretical analysis and experiments show that the autoencoder has similarity and can be used to quickly filter tags that do not need to be compared. Compared with the local sensitive hash algorithm, autoencoder has randomness and higher security. The method of this paper can not only substantially improve the efficiency of R-MLE-based secure deduplication method, but also introduce artificial intelligence into secure deduplication research, which is expected to become a new research hotspot.