Discrminative Geometry Preserving Projections

Dimension reduction algorithms have attracted a lot of attentions in face recognition and human gait recognition because they can select a subset of effective and efficient discriminative features. In this paper, we apply the discriminative geometry preserving projections (DGPP), a new subspace learning algorithm to address these problems. DGPP models both the intraclass geometry and interclass discrimination. Meanwhile, DGPP will not meet the undersampled problem. Thoroughly empirical studies on YALE face database, UMIST face database, FERET face database and USF human-ID gait database demonstrate that DGPP is superior the popular algorithms for dimension reduction, e.g., PCA, LDA, NPE and LPP.