28 November 2022 in Computers & Security
26 October 2022 in Computers & Security
18 March 2023 in Knowledge and Information Systems
9 May 2023 in Computers & Security
28 December 2024 in Future Generation Computer Systems
9 June 2025 in IEEE Transactions on Information Forensics and Security
Trajectory publication under local differential privacy (LDP) has recently become a research focus. Existing solutions rely on geospatial discretization, elevating trajectory description granularity to alleviate noise injection. However, discretization itself also leads to trajectory information loss. Determining discretization granularity to balance differential noise and trajectory accuracy is challenging. Besides, these solutions commonly protect trajectory privacy via whole-region perturbation yet ignore the actual reachable range of trajectories, resulting in impractical published trajectories. To address these issues, we propose LDPTP, a novel LDP-based trajectory publication method that achieves high-quality publication by perceiving regional popularity in a privacy-preserving way. Specifically, we design a privacy-accuracy balancing mechanism for discretization granularity selection, which can effectively measure the impact of different granularities on noise error and information loss through the Bernoulli model and information entropy, enabling optimized discrete trajectories acquisition. Furthermore, a regional popularity-based perturbation method is presented, which utilizes trajectory distribution features to capture popular regions and then combines region similarity to generate private mobility patterns that better preserve trajectory utility. Finally, we devise a transition probability correction method to enhance the accuracy of Markov model learned from these private patterns, realizing high-utility trajectory synthesis for publication. Extensive experiments are conducted on real-world and synthetic datasets under three levels of utility metrics. The results demonstrate that our proposed LDPTP significantly outperforms the baseline methods.