Transform domain based visual and infrared image fusion method is an important research direction. All kinds of natural images could not be expressed effectively by wavelet transform with only one kind of wavelet basis functions due to the high redundancies of its linear and curve singularity expression. Multi-resolution singular value decomposition (MR-SVD) computed the transformation matrix from the original image. With the computed transformation matrix, the original image is decomposed to unrelated “smooth” and the “detail” components. On each layer of the smooth components, the singular value decomposition (SVD) is used to replace the wavelet filter, realizing the multi-level decomposition. A novel visual and infrared image fusion algorithm is presented because of the better sparsity and adaptability of multi-resolution singular value decomposition (MR-SVD), which could resolve the difficult problem of wavelet function basis selection for different kind of visual and infrared images. The same transformation matrixes computed from original visual or infrared imagery used to decompose the original images with MR-SVD, which could reduce the blurring problem of fusion image got by the average transformation matrixes. Then, cycle spinning is employed to remove the artifacts in the fusion image. experimental results according to both the subjective and objective criteria, including the average, standard deviation and average MI, indicate that the proposed method could get better fusion results compared to methods like wavelet transform.