Synchronized ReID with Expanded Cross Neighborhood Re ranking

Abstract

In view of person re-identification (re-ID) as a retrieval process, re-ranking is a crucial step to improve its performance. In the re-ID research, limited effort has been devoted to re-ranking, especially when it comes to fully automatic, unsupervised solutions. In this paper, we propose enhanced expanded cross-neighborhood based Re_ranking with Synchronized ReID in which global features are extracted which are mutually learned with local features and then re ranked to improve performance. Enhanced ECN greatly improves the person retrieval method. Global feature learning greatly took advantage from local feature learning, which performs a synchronization/alignment without requiring extra monitoring by calculating the shortest path between two sets of local features. After the joint learning, we only match the global feature to measure the similarities between images and effective re ranking is applied in the test set to greatly improve the performance of the ReID system.