Summary: Aimed at the defect that the local characteristics are hard to be well represented in spatial domain, the training set is directively transformed into wavelet domain. In training phase, the K-SVD dictionary learning algorithm is used to train four pairs of dictionaries of subband high-low resolution respectively for high-low resolution feature in extracted wavelet domain, and the subband dictionaries obtained are used for high-resolution image reconstruction in wavelet domain. In order to improve the quality of the reconstructed image, an adaptive mixed sample ridge regression (AMSRR) model is proposed to modulate the high-frequency component of the image. Massive experimental results show that the proposed algorithm is superior to the competitive spatial domain methods both in visual effect and quantification index (PSNR and SSIM).