<p>Visible-Infrared person re-identification (VI-ReID) aims to match pedestrian images captured by visible and infrared cameras, which is essential for round-the-clock surveillance and cross-spectrum person retrieval. However, VI-ReID remains a challenging task due to the significant modality gap and the difficulty in learning modality-invariant yet discriminative representations. In this paper, we propose a hybrid attentive feature alignment framework for VI-ReID that addresses the challenges of modality discrepancy and representation discrimination. Specifically, a dual-stream network is adopted to extract modality-specific features from visible and infrared images. To enhance intra-modality discrimination, we introduce a local spatial attention module that captures salient local patterns within each modality. Meanwhile, a global semantic attention module is designed to model high-level semantic dependencies across modalities, enabling robust cross-modal semantic understanding. The outputs of both attention modules are fused and further refined through a cross-modal feature alignment mechanism, which reduces distribution gaps and promotes modality-invariant feature learning. Extensive experiments on two public VI-ReID datasets demonstrate that our method achieves superior performance compared to existing approaches, validating its effectiveness and generalizability.</p>