The nonsubsampled contourlet transform(NSCT) is based on a nonsubsampled pyramid structure and nonsubsampled directional filter banks.NSCT is shift-invariant ,of high redundancy compared with contourlet transform ,and it can overcome Pseudo-Gibbs phenomenon. An image can be decomposed into multiscale and multidirectional details.These details represent image characteristics in different directions and different frequency bands,which simplify the relationship between coefficients. Learning-based super-resolution reconstruction algorithm has predictability of a whole image.Learning-based method combined with NSCT can improve training accuracy. Experiments with fingerprint image show that the algorithm has good performance and the reconstructed image can maintain the feature points possessed of the high resolution image and is closed to original image with good texture details.