计算机工程 ›› 2019, Vol. 45 ›› Issue (5): 194-198.doi: 10.19678/j.issn.1000-3428.0050309

• 图形图像处理 • 上一篇    下一篇

锚点领域回归与稀疏表示的图像超分辨率方法

端木春江,左德遥   

  1. 浙江师范大学 数理与信息工程学院,浙江 金华 321004
  • 收稿日期:2018-01-26 出版日期:2019-05-15 发布日期:2019-05-15
  • 作者简介:端木春江(1974—),男,副教授,主研方向为图像处理、视频通信;左德遥,硕士研究生。
  • 基金项目:

    浙江省自然科学基金(LY15F010007,LY18F010017)。

Image Super-resolution Method via Anchored Neighborhood Regression and Sparse Representation

DUANMU Chunjiang,ZUO Deyao   

  1. College of Mathematics,Physics and Information Engineering,Zhejiang Normal University,Jinhua,Zhejiang 321004,China
  • Received:2018-01-26 Online:2019-05-15 Published:2019-05-15

摘要:

结合锚点领域回归与稀疏表示方法,提出一种改进的图像超分辨率方法。通过对高分辨率图像采用模糊和下采样操作生成低分辨率图像,基于锚点邻域回归的线性映射函数训练投影矩阵,利用稀疏表示的方法训练和学习稀疏字典对。在图像放大阶段,根据训练好的投影矩阵重建主要高频特征,利用稀疏字典对补充残差高频特征。实验结果表明,该方法能较好地保持图像的局部细节信息,减少块效应和伪影效应。

关键词: 图像超分辨率, 字典学习, 稀疏表示, 锚点邻域回归, 图像放大

Abstract:

Combining anchored neighborhood regression and sparse representation methods,this paper proposes an image super-resolution method.By blurring and subsampling high-resolution image to generate low-resolution image,the linear mapping function based on anchored neighborhood regression is used to train the projection matrix,and sparse representation is used to train and learn sparse dictionary pairs.In the online image magnification stage,the main high frequency features are generated by using the trained projection matrix.Then,the sparse dictionary pairs are employed to reconstruct the residual high frequency features.Experimental results show that the proposed method can maintain the local detail information of the image,reduce the blocks and aliasing artifacts.

Key words: image super-resolution, dictionary learning, sparse representation, anchored neighborhood regression, image magnification

中图分类号: