Image Super-Resolution Reconstruction Based on Sparse Dictionary Learning and non-Local Self-Similarity

Image super-resolution reconstruction plays an important role in computer vision systems. However, it is still a challenging work, especially in enhancing the high-resolution (HR) image quality and validating the algorithms. This paper proposes a super-resolution reconstruction method based on sparse dictionary learning and non-local self-similarity. Firstly, we assume that HR and low-resolution (LR) image blocks have the same sparse representation coefficients. The LR dictionary and sparse representation coefficients are obtained by using the K-SVD algorithm. Then an intermediate HR result is reconstructed by utilizing the HR training image feature blocks. Finally, the non-local similarity regularization is performed to estimate the final HR image to eliminate the artificial traces of the intermediate result which further improves the reconstruction effect. Experimental results demonstrate that the proposed method can better restore the edge details and improve the quality of HR image compared with the existing methods.