Couple manifold discriminant analysis with bipartite graph embedding for low-resolution face recognition

This brief paper presents a novel method for low-resolution face recognition. We introduce a generalized bipartite graph to discretely approximate the underlying manifold structure of face sets with different resolutions. Unlike traditional graph-based methods that only construct the graph based on one sample set, the proposed method constructs the generalized bipartite graph on two heterogeneous sample sets and contains more completed information. Our method learns a couple of mappings that project the face sets with different dimensions into a unified feature space which favors the task of classification. Specifically, in this unified space, our method preserves within-class local geometrical structure according to the network topology of the generalized bipartite graph and maximizes between-class separability at the same time. Experimental results on two benchmark face databases demonstrate the effectiveness of our proposed algorithm.