Analyzing ameliorated nonnegative matrix factorization for wood image representation

Nonnegative matrix factorization (NMF) is an unsupervised method whose aim is to find an approximate factorization V_n*m = W_n*r*H_r*m into non-negative matrices W_n*r and H_r*m. This paper presents an extension to NMF and discusses the development and the use of damped Newton based the non-negative matrix factorization called DNNMF with good convergence properties for wood image representation by adding a diagonal correction to the stiffness matrix and employing a Newton direction in the line search until any constraints become active. This method can make sure the convergence of the cost functions and has been tested with color images based on the LBP feature histograms extracted by Local Binary Pattern (LBP) from the feature subspaces structured by DNNMF. Comparative experiments show that the proposed method is effectual and practical with good research values and potential applications.