In the field of image processing, dictionary learning and the mapping from Low-Resolution (LR) image to High-Resolution(HR) image are two important components of image Super-Resolution(SR) algorithms based on sparse representation theory. Due to the rich and varied image types, a single dictionary does not represent the image very well. And the strict equaling to the mapping between LR and HR images also limits the image reconstruction effect. From above two aspects, more inclusive multi-dictionary and the more relaxed coupled dictionary sparse learning algorithms are adopted to perform SR reconstruction of the image. First of all, the method performs multiple adaptive clustering on the basis of non-local self similarity of images in this paper. Secondly, the optimal clustering is selected, and the dictionary is got by the coupling dictionary sparse learning algorithm. Finally, the input LR images are classified and reconstructed to obtain HR images. The experimental results show that Peak Signal to Noise Ratios(PSNRs) of image Leaves, Barbara and Room with the proposed clustering algorithm is higher than that with original sparse learning algorithm by 0.51 dB, 0.21 dB, 0.15 dB respectively.