Two Dimensional Nonnegative Matrix Factorization

Nonnegative Matrix Factorization (NMF) has been widely used in computer vision and pattern recognition. It aims to find two nonnegative matrices whose product can well approximate the original matrix, which naturally leads to parts-based representation. In this paper, we propose a Two Dimensional Nonnegative Matrix Factorization (2DNMF), specifically for a sequence of matrices. In contrast to NMF which applies for only a single matrix, and finds only one base matrix, 2DNMF aims to find two base matrices to represent the input matrices in a low dimensional matrix subspace. It not only inherits the advantages of NMF, but also owns the properties low computational complexity, as well as high recognition accuracy. Experiments on benchmark image recognition data sets illustrate that the proposed method is very effective and efficient.