This paper presents an unsupervised salience learning approach for person re-identification, addressing the challenge of recognizing individuals across different camera views. The method extracts distinctive features without requiring identity labels during training, making it adaptable to various camera settings. Key contributions include: 1. **Dense Correspondence**: Utilizes patch matching with adjacency constraints to handle misalignment caused by viewpoint changes and pose variations. 2. **Unsupervised Salience Learning**: Learns human salience in an unsupervised manner, enhancing the discriminative power of patch matching. 3. **Bi-directional Weighted Matching**: Incorporates salience information into the matching process to improve accuracy. The effectiveness of the approach is validated on the VIPeR and ETHZ datasets, demonstrating superior performance compared to existing supervised and unsupervised methods. The results show that the proposed method significantly improves the rank-1 matching rate, achieving around 26% on the VIPeR dataset and outperforming other methods on the ETHZ dataset.This paper presents an unsupervised salience learning approach for person re-identification, addressing the challenge of recognizing individuals across different camera views. The method extracts distinctive features without requiring identity labels during training, making it adaptable to various camera settings. Key contributions include: 1. **Dense Correspondence**: Utilizes patch matching with adjacency constraints to handle misalignment caused by viewpoint changes and pose variations. 2. **Unsupervised Salience Learning**: Learns human salience in an unsupervised manner, enhancing the discriminative power of patch matching. 3. **Bi-directional Weighted Matching**: Incorporates salience information into the matching process to improve accuracy. The effectiveness of the approach is validated on the VIPeR and ETHZ datasets, demonstrating superior performance compared to existing supervised and unsupervised methods. The results show that the proposed method significantly improves the rank-1 matching rate, achieving around 26% on the VIPeR dataset and outperforming other methods on the ETHZ dataset.