非局部联合稀疏近似的超分辨率重建算法

Super-resolution Reconstruction Algorithm Based on Non-local Simultaneous Sparse Approximation

该文结合联合稀疏近似和非局部自相似的概念,提出非局部联合稀疏近似的超分辨率重建方法。该方法将输入图像的跨尺度高、低分辨率图像块统一进行联合稀疏编码,建立它们之间的稀疏关联,并将这种关联作为先验知识来指导图像的超分辨率重建。该文方法保证跨尺度自相似集具有相同的稀疏性模式,能更有效地利用图像的自相似性先验信息,提高算法的自适应性。通过自然图像实验,与其它几种基于学习的超分辨率算法对比,超分辨率效果有较好改善。

A novel super-resolution reconstruction method based on non-local simultaneous sparse approximation is presented, which combines simultaneous sparse approximation method and non-local self-similarity. The sparse association between high- and low-resolution patches pairs of cross-scale self-similar sets via simultaneous sparse coding is defined, and the association as a priori knowledge is used for super-resolution reconstruction. This method keeps the patches pairs the same sparsity patterns, and makes efficiently use of the self-similar information. The adaptability is enhanced. Several experiments using nature images show that the presented method outperforms other several learning-based super-resolution methods.