An infrared and low-level-light image fusion algorithm based on image sparse representation is proposed according to human visual systems and the over-complete sparse representation. In the method, an image is segmented into partly overlapped image patches firstly. The image patches are decomposed by an orthogonal matching pursuit algorithm. Then, the maximum fusion rule is used to choose suitable fusion coefficients for the reconstruction of image patches. Thus, the fused images are obtained. The experimental result shows that compared with the traditional fusion methods such as wavelet transform, Laplacian pyramid and principal component analysis methods, the proposed method has better fusion effectiveness.