A novel matching method based on Robust Nonlinear Projective Nonnegative Matrix Factorization (RNPNMF) is proposed to find the correspondence among different images containing the same object. We show how the features point-sets can be matched using their common projection space. The contribution can be divided into two parts. Firstly, a robust RNPNMF method is developed to capture the common projection space of the feature point-sets. Secondly, a matching approach is derived from the projections on the common projection space of the feature point-sets. Finally, two experiments are conducted to verify the effectiveness of the proposed method. The experimental results show that compared with the existing method, our method is more effective in matching the feature point-sets and can be generalized well to registration.