Unsupervised person re-identification via graph-structured image matching

Abstract

This paper presents a novel unsupervised framework to solve both person re-identification and partial person re-identification problems. For each pedestrian image, we first use an established image segmentation method to generate superpixels to construct an Attributed Region Adjacency Graph (ARAG) in which nodes corresponding with superpixels and edges representing correlations between superpixels. We then apply region-based Normalized Cut to the graph to merge similar neighbouring superpixels to form natural image regions corresponding to various body parts and backgrounds. To tackle the occlusion problem often encountered in these applications, we apply Denoising Autoencoder with nonnegativity constrains to learn robust and part-based representation of image patches in each node of the graph. Finally, the similarity of an image pair is measured by the Earth Mover’s Distance (EMD) between the robust image signatures of the nodes in the corresponding ARAGs. We evaluate our methods on both person re-id and partial person re-id datasets and the results show that our new framework outperforms state of the art methods.