An efficient hierarchical multi-scale image fusion method was proposed. Several polarization parametric images were fused for enhancement by non-negative matrix factorization. A polarization parametric fusion image was obtained. This image contained complete polarization information with low redundancy. Then, with two-dimensional empirical mode decomposition, polarization parameters fusion image and polarization intensity image were decomposed. High-frequency component and low-frequency component were fused by weighted average function. In order to determine fusion weight, exhaustive search method was taken. Finally, the given results of high-frequency sub-images and low-frequency sub-images were inverse transformed. The experiments demonstrate that the proposed fusion method has the ability to enhance the details and contrast of polarization image.