During image super-resolution reconstruction for multi-dictionary learning, common methods such as K-means clustering, Gauss mixed model clustering and so on can lead to poor quality and instability of image reconstruction. To solve the problem, we propose a novel image super-resolution reconstruction algorithm based on hierarchical clustering. Firstly, features are extracted from sample image blocks, and hierarchical clustering is performed, then K dictionaries are trained with improved principal component analysis method. Secondly, the test images are cut into a number of image blocks, and the most suitable dictionary is adaptively matched to reconstruct the image block. Finally, the whole image is optimized to achieve global reconstruction. The results show that the proposed algorithm in this paper has high feasibility, and can effectively improve the reconstruction quality of image. Compared with peak signal-to-noise ratio and structural similarity of the images reconstructed by the traditional algorithms, those of the images reconstructed by the proposed algorithm increase.