As traditional multi-focus image fusion methods can not effectively measure the partitioning focus regions in images, a novel algorithm by using super-resolution image reconstruction for multi-focus image fusion was proposed to solve the problem. The algorithm measured the in-focus and out-of-focus regions and performed the super-resolution image reconstruction for the clear area with sparse representation. Firstly, the spatial frequency method was used to extract the in-focus and out-of-focus regions in source images. Then, the main-clear and sub-clear parts within in-focus regions were identified and their real down-sampling scales for each part were calculated. Finally, the sub-clear parts were reconstructed in super-resolution through learning multi-scale sparse dictionaries and the fused image was obtained by combining the different parts of source images. The experimental results show that the proposed method can provide clear images and better facus performance. As compared with the conventional methods, such as Harr wavelet, Nonsubsampled Contourlet Transform (NSCT), and shearlet transform,the proposed method enhances its Entropy (EN) and Peak Signal-to-Noise Ratio (PSNR) by 1%,and 0.62dB, respectively, the clarity (SP) and spatial frequency (SF) by 30%, and the Mean Square Error (MSE) is decreased by about 6%.