Person re-identification by using an iterative sparse ranking method

Abstract

Based on sparse basis expansions of targets, the discriminative which in terms of a labelled gallery of known individuals is proposed as a method for person re-identification. It is capable of ranking many candidate targets by using proposed iterative extension. The approach make sure that the best candidates are ranked at each iteration and redistribute energy among the most relevant contributing elements by using soft and hard weighting. This approach also influences a novel visual descriptor which show to be discriminative while remaining robust to pose and illumination variations. By demonstrating this approach the state-of-the-art performance are achieved on single- and multi-shot person re-identification scenarios on the VIPeR, CAVIAR4REID, ETHZ, and i-LIDS datasets by using the extensive comparative evaluation. The state-of-the-art rank-1 performance are improved by 6 percentage points on VIPeR and whereas by 20 on CAVIAR4REID by the combination of iterative sparse basis expansion and descriptor with a single gallery image per person when compared to other methods. It also improves the state-of-the-art by 17 percentage on i-LIDS and by 72 on CAVIAR4REID at rank-1 by multiple gallery and probe images per person. At about 30 re-identifications per second, the approach is capable of single-shot person re-identification while containing hundreds of individuals over galleries and it is also efficient.