With the growing prevalence of video surveillance systems, person re-identification technology has become increasingly important for public safety. However, existing technologies face the challenge of balancing privacy protection and model performance. To address these issues, this paper proposes a federated privacy-preserving framework named FPPReID. Firstly, it embeds frequency domain segmentation into federated learning to retain data locally and protect privacy. Secondly, it designs an information compensation and attention correction mechanism to enhance model robustness. By generating feature masks from low-frequency information and optimizing attention allocation using human keypoint detection, this mechanism balances privacy protection and model performance. Experimental results on the Market1501 dataset demonstrate that the proposed methods achieve excellent performance in both privacy protection metrics, such as PSNR and SSIM, and re-identification performance such as Rank-1 accuracy. These findings indicate that the proposed methods effectively resolve the trade-off between privacy protection and model performance, offering new insights for the development of privacy-preserving person re-identification technology.