A Secure Deduplication Method Based on the Transformer Model

Random Message Locked Encryption (R-MLE) is an effective scheme for secure deduplication in cloud storage. However, due to its reliance on bilinear mapping, the fingerprint comparison process in deduplication incurs high computational costs. To address this, we propose a secure deduplication method based on a Transformer encoder. The Transformer encoder extracts feature vectors from the data, and a pooling operation converts these high-dimensional vectors into fixed-length low-dimensional vectors as Feature Tags. By comparing the similarity between Feature Tags, the proposed method achieves efficient filtering of fingerprint tags, thereby reducing computational overhead. Experimental results demonstrate that our method generates Feature Tags faster, ensures a more uniform distribution, and improves similarity comparison performance, ultimately enhancing deduplication efficiency.