Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (5): 197-202.DOI: 10.3778/j.issn.1002-8331.1711-0297

Previous Articles     Next Articles

Single Image Super-Resolution by Multi-Dictionary Learning

WANG Chong, SHANG Xiaoqing   

  1. School of Mathematics and Statistics, Xidian University, Xi’an 710126, China
  • Online:2019-03-01 Published:2019-03-06


王  冲,尚晓清   

  1. 西安电子科技大学 数学与统计学院,西安 710126

Abstract: This paper presents a new multi-dictionary learning algorithm for the problem of single super-resolution for the common dictionary-based SR algorithm ignoring the structure of the image, and some high-frequency information in the image reconstruction process. Firstly, the dominant direction of each image patch should be computed before clustering, then training dictionaries. Next, the residual image patches which computed by the dictionaries can be used to get residual dictionaries. Last, adjusted anchored neighborhood regression is used to reconstruct image, experimental results show that the algorithm has a better result both on visual effect and objective compared with other methods.

Key words: super-resolution, anchored neighborhood regression, image reconstruction, residual

摘要: 对于目前图像超分辨率重建算法中的问题,忽略重建图像结构性和重建过程中丢失高频信息,提出了一种基于多字典的单幅图像超分辨率重建算法。在字典学习阶段根据每个图像块的主方向角,对所有训练图像块进行聚类并训练各类的字典。利用训练得到的字典重建训练样本并计算各类的残差图像块,然后对残差图像块再进行聚类、训练残差字典。用锚定邻域回归方法重建高分辨率图像,实验结果表明,该算法在客观评价和视觉效果上均优于许多优秀的图像超分辨算法。

关键词: 超分辨率, 锚定邻域回归, 图像重建, 残差