Paper accepted @ IEEE Transactions on Multimedia

Title:  Indistinguishability Analysis of JPEG Image Encryption Schemes

Authors: Yuan Yuan,  Lingfeng Qu, Ji Zhang, Ningxiong Mao, Hadi Amirpour

Abstract: JPEG images are widely used for communication and storage, making secure encryption essential for privacy protection. Existing JPEG encryption studies primarily rely on empirical metrics such as visual distortion, key space, or correlation, while overlooking the formal indistinguishability against chosen-plaintext attacks (IND-CPA) property. This work provides the first systematic analysis of existing JPEG encryption schemes from the IND-CPA perspective. A new metric, termed feature change rate, is introduced to quantify the preservation of residual features. Furthermore, the relationship between feature change rate and key estimation success under CPA is established, indicating that smaller feature changes result in higher attack accuracy. Based on these findings, we propose a set of design principles for constructing secure and practical JPEG encryption schemes. Finally, we outline a feature-changing encryption strategy that enhances IND-CPA security while maintaining JPEG compatibility and compression efficiency.