Graph Embedding Projective Non-negative Matrix Factorization Method for Image Feature Extraction
Wang Jua · 2014
To overcome the disadvantage that projective non-negative matrix factorization(PNMF)fails to discover the intrinsic geometrical and discriminating structure,a novel graph embedding projective non-negative matrix factorization(GEPNMF)was proposed for image feature extraction.The paper constructed two adjacent graphs that are separately used to characterize the intrinsic geometrical structure of data and interclass separability.Using the Laplacian matrices of the adjacent graphs,the paper designed a graph embedding regularization that incorporates with PNMF's objective function to construct the GEPNMF's objective function.Since the graph embedding regularization is adopted by the objective function,the learned subspace of GEPNMF can preserve the data geometrical structure while it maximizes the margins between different classes.That is to say,it has more discriminability.In addition,the paper introduced an orthogonal regularization into the objective function to ensure the learned bases to be parts-based.The paper deduced a multiplicative update rule(MUR)to optimize the objective function.The experimental results on Yale and CMU PIE face image datasets suggest the effectiveness of GEPNMF.