Super-Resolution Person Re-Identification With Semi-Coupled Low-Rank Discriminant Dictionary Learning

• DOI: 10.1109/tip.2017.2651364
• Publication Date: 2017-01-10T19:25:38Z
• Authors: Xiao-Yuan Jing; Xiaoke Zhu; Fei Wu; Ruimin Hu; Xinge You; Yunhong Wang; Hui Feng; Jing-Yu Yang
• Keywords: Discriminative model; Regularization; Identification; Feature (linguistics); Rank (graph theory)

Person re-identification has been widely studied due to its importance in surveillance and forensics applications. In practice, gallery images are high resolution (HR), while probe usually low (LR) the identification scenarios with large variation of illumination, weather, or quality cameras. this kind scenarios, which we call super-resolution (SR) person re-identification, not well studied. paper, propose a semi-coupled low-rank discriminant dictionary learning (SLD2L) approach for SR task. With HR LR pair mapping matrices learned from features training images, SLD2L can convert into features. To ensure that converted have favorable discriminative capability dictionaries characterize intrinsic feature spaces design term regularization SLD2L. Moreover, considering results different degrees loss types visual appearance features, multi-view (MVSLD2L) approach, learn type-specific mappings each type feature. Experimental on multiple publicly available data sets demonstrate effectiveness our proposed approaches