«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:

42

期数:

2021年03期

页码:

59

栏目:

出版日期:

2021-05-10

文章信息/Info

Title:

Semi-coupled dictionary learning super-resolution reconstruction model introducing detail constraint factors

作者:

黄宇达;  王迤冉 ;  牛四杰

周口职业技术学院信息工程学院;周口师范学院网络工程学院;济南大学信息科学与工程学院;

Author(s):

Huang Yuda; Wang Yanran; Niu Sijie;

College of Information Engineering of Zhoukou Vocational and Technical College; School of Network Engineering, Zhoukou Teachers College; School of Information Science and Engineering, Jinan University;

关键词:

超分辨率; 半耦合字典学习; 细节约束因子; 拉普拉斯分布

Keywords:

Super resolution;  semi -coupled dictionary learning;  detail constraint factor;  Laplas distribution

DOI:

10.13705/j.issn.1671-6833.2021.03.010

文献标志码:

A

摘要:

为了提升单幅图像的超分辨率重构细节,提出了一种基于细节保持的超分辨率重构方法针对半耦合字典学习超分辨率算法细节保持不够高的缺陷,采用细节约束因子与半耦合字典交替学习策略在重构阶段,利用图像水平方向与垂直方向的梯度构建细节约束因子,并引入到半耦合字典学习框架,进一步提高重构精度为了改进细节约束因子在重构过程中的贡献度,采用边界差异的拉普拉斯分布实现参数的自适应选择相比于半耦合字典学习超分辨率算法,该方法在峰值信噪比方面平均提升1.5%实验结果表明,该方法在主观和客观评价标准下均取得了较好的重构效果,提升了超分辨率重构质量

Abstract:

In order to improve the super-resolution reconstruction quality of single image，an improved learning based super-resolution approach was proposed in this paper. To tackle the problem of low details of semi-coupled dictionary learning super-resolution algorithm ， the paper presented learning strategy where detail constraint factor and semi-coupled dictionary learning were performed in turn. In reconstruction stage，detail constraint factor was designed by the gradient in both horizontal and vertical direction .Combined with semi-coupled dictionary learning ，detail constraint factor was used to further improve the super-resolution reconstruction quality. In order to improve the contribution of detail constraint factor on preserving boundary information，the adaptive regular parameter was explored via the approximate Laplacian distribution of edge difference .Compared with the semi coupled dictionary learning super-resolution algorithm，the peak signal-to-noise ratio of this method was increased by 1.5% on average . Experiments demonstrated that the proposed method could achieve better reconstruction effect in both subjective and objective evaluation and improve the quality of super-resolution.