引入细节约束因子的半耦合字典学习超分辨率重构模型  

Semi-coupled Dictionary Learning Super-resolution Reconstruction Model with Detail Constraint Factor

作　　者：黄宇达[1] 王迤冉 牛四杰[3] HUANG Yuda;WANG Yiran;NIU Sijie(College of Information and Engineering,Zhoukou Vocational and Technical College,Zhoukou 466000,China;College of Network Engineering,Zhoukou Normal University,Zhoukou 466000,China;College of Information Science&Engineering,Jinan University,Jinan 250000,China)

机构地区：[1]周口职业技术学院信息工程学院,河南周口466000 [2]周口师范学院网络工程学院,河南周口466000 [3]济南大学信息科学与工程学院,山东济南250000

出　　处：《郑州大学学报（工学版）》2021年第3期59-64,共6页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(61103143);河南省科技计划项目(112300410307)。

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。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 semicoupled 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.

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 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]