利用稀疏编码的遥感影像超分辨率重建  

Sparse Coding Based Remote Sensing Image Super Resolution Reconstruction

作　　者：杨轶斐 刘士彬[1] 戴芹[1] 马勇[1,2] 何阳[1,2] YANG Yifei LIU Shibin DAI Qin MA Yong(Institute of Remote sensing and Digital Earth ,Chinese Academy of Sciences ,Beijing 100094 ,China University of Chinese Academy of Sciences ,Beijing 100049 ,China)

机构地区：[1]中国科学院遥感与数字地球研究所,北京100094 [2]中国科学院大学,北京100049

出　　处：《遥感信息》2017年第2期31-37,共7页Remote Sensing Information

基　　金：中国科学院遥感与数字地球研究所"一三五规划"项目(Y3ZZ07101A)

摘　　要：针对现有的超分辨率重建算法存在重建效果不够理想以及算法效率低下的问题,提出了一种利用稀疏编码的超分辨率重建算法,利用遥感图像在特定字典下具有稀疏性,通过学习遥感图像集获取的字典,在已有遥感影像的基础上获取更高分辨率的影像,从而便于影像视觉解译和信息提取。所提出方法建立了高低分辨率影像块稀疏系数间的非线性映射关系,极大地减少了字典模型的冗余和误差,提高了重建的效率。实验表明,该算法不仅在信噪比值和结构相似性值优于其他算法,还提高了算法效率。This paper presents a new sparse coding based super resolution reconstruction（SCSRR）algorithm in order to obtain better effects and higher efficiency than the existing super-resolution reconstruction algorithms.The proposed SCSRR algorithm makes use of the sparsity of remote sensing images in specific dictionaries,and reconstructs high resolution images through a set of sparse coefficients after learning dictionaries by the existing remote sensing images,which is convenient for image visual interpretation and information extraction.This algorithm sets up a nonlinear mapping between low and high resolution patch-pairs＇ sparse coefficients,greatly removes redundancies and errors in dictionaries,and improves the efficiency of super resolution reconstruction.Experimental results show that the proposed SCSRR algorithm can not only outperform others in PSNR value and SSIM value,but also greatly reduce the reconstruction time.

关 键 词：超分辨率重建 稀疏编码 字典学习 高低分辨率 非线性映射 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]