Information Disclosure Risk of Thumbnail-Preserving Encryption

With the rapid growth of cloud services, the storage of images in cloud environments requires secure and effective data encryption methods. Many thumbnail-preserving encryption (TPE) methods have thus been proposed to balance privacy and usability of image data. However, the exposure of thumbnail information in TPE methods may introduce privacy leakage risk, and a systematic evaluation of their security has not yet been conducted. In this paper, we propose a new Mamba-Transformer cooperation Network (MTNet) to recover the original images from the limited exposed thumbnail information, highlighting the information disclosure problem in TPE. Specifically, the core model component integrates a Mamba block and a Transformer block, which employ the powerful capabilities of the Mamba for wide field dependency modeling and the Transformer for effective channel interaction. Besides, the cascade architecture incorporates an intermediate output that provides supplementary information and achieves multilevel supervision, thereby improving the quality of the final output. Finally, to better utilize the subtle details in different levels, we propose a multi-scale fusion module that adaptively integrates features from various stages of the encoding process. The experimental results achieved by our proposed MTNet reveal that the privacy risk associated with TPE is significantly underestimated and more robust defense mechanisms are required. Source code is available at https://github.com/HITLiXincodes/MTNet.