Improved Person Re-identification Method Based on Partial-Global Context Network

Person re-identification (Re-ID) focuses on identifying target individuals from images captured by multiple cameras. However, traditional Re-ID methods encounter issues such as non-salient discriminative features and insufficient expression of key pedestrian features. In this paper, a novel Re-ID method based on a Partial-Global Context Network is proposed to address these challenges. The Partial Relation Network is designed to explore non-salient discriminative features, while the Global Context Network enables the model to focus on effective feature regions with varying degrees of saliency, thereby reducing intra-class differences among pedestrians and enhancing the discriminative final feature representation. The proposed method has been experimentally validated on the Market1501 and DukeMTMC-ReID datasets. Compared with some other existing method RRID, the mean Average Precision (mAP) and rank-1 accuracy have been improved by 1.3% and 0.5%, respectively.