Context-based Matching Refinement for Person Search

Byeong-Ju Han, Jae-Won Yang, Oggyu Lee, Jae-Young Sim

Abstract: Person search is the combined task of person detection and re-identification that aims to find target persons having the same identity to a given query person. Existing methods of person search mainly focus to extract useful features of persons, but do not use the relationship among the persons in gallery images during the test phase. In this paper, we propose a context-based matching refinement scheme for person search by utilizing the contextual relationship of persons in gallery images. We compute the matching context such that a target person has a high context value with respect to a query person, when it has relatively high similarities to the positive samples of the query person. Experimental results show that the proposed method significantly outperforms the existing state-of-the-art person search methods on PRW dataset.