Image Super-Resolution via Deep Dictionary Learning.

The method of image super-resolution reconstruction through a dictionary usually only uses a single-layer dictionary, which not only fails to extract the deep features of the image, but also the trained dictionary may be relatively large. This paper proposes a new deep dictionary learning model. First, after preprocessing the images of the training set, the dictionary is trained by the deep dictionary learning method, and the super-resolution reconstruction is performed by adjusting the anchored neighborhood regression method. The proposed algorithm is compared with several classical algorithms on the Set5 data set and Set14 data set. The visualization and quantification results show that the proposed algorithm has a good improvement in PSNR and SSIM compared with the traditional super-resolution algorithm, and effectively reduces the dictionary size and saves reconstruction time.