Person Re-Identification via Multi-Dimensional Attention Mechanisms
Jiayu Zou, Shuren Zhou, Xinlan Duan · 2025
The person re-identification task aimss to match images of the same individual from different camera views and has widespread applications in real-world scenarios. However, existing methods cannot effectively remove background information and do not pay enough attention to key information areas, resulting in insufficient accuracy. To address these issues, we propose an improved ReID method, named the Multi-Dimensional Attention Network (MDAN), which integrates multi-scale attention, spatial attention, and channel attention mechanisms. The multi-scale attention module captures human features at different scales, enhancing adaptability to scale variations. The spatial attention module guides the model to focus on key body regions, mitigating background interference. Meanwhile, the channel attention module optimizes the weighting of feature channels, improving the model’s discriminative capacity. Experiments on multiple public datasets show that the proposed method achieves significant improvements in accuracy over existing approaches, verifying its effectiveness and robustness.