Abstract: In order to solve the problem of high complexity and poor reconstruction effect of traditional reconstruction methods and make the super-resolution reconstruction images have good clarity, a new super-resolution reconstruction algorithm based on learning is studied based on sparse representation principle. The super-resolution optimal solution is realized. By inputting a low-resolution image, the corresponding image feature block can be obtained by bicubic interpolation. A set of sparse representation coefficients can be obtained in the low resolution part of the dictionary (DL). According to the set of sparse representation coefficients, the high resolution image blocks are reconstructed by reverse projection on the high resolution dictionary (Dh), and the complete high resolution images are obtained by combining them together. The simulation results show that compared with the bicubic interpolation method, the reconstructed image of this algorithm has better quality.