Grassmannian Manifolds Discriminant Analysis Based on Low-Rank Representation for Image Set Matching

Abstract

Recently, a discriminant analysis approach on Grassmannian manifolds based on a graph-embedding framework is proposed for image set matching. However, its accuracy critically depends on the number of local neighbours when constructing a similarity graph. In this letter, a novel approach with fixed neighbour numbers is presented to implement graph embedding Grassmannian discriminant analysis. The approach utilizes the ‘low-rank component’ of set to represent each image set. During the manifold mapping, the nearest neighbour structure of nodes with same label and all the different label information are employed to preserve the local geometrical structure. Experiments on two image datasets (15-scenes categories and Caltech101) show that the proposed method outperforms state-of-the-art methods for image sets matching.