利用光流配准进行嫦娥一号CCD多视影像超分辨率重建  

Optical Flow Based Super-resolution of Chang'E-1 CCD Multi-view Images

作　　者：陈王丽[1] 孙涛[2] 陈喆[3] 马国锐[1] 秦前清[1] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079 [2]武汉大学电子信息学院,湖北武汉430072 [3]中国电子科技集团公司第十五研究所,北京100083

出　　处：《武汉大学学报（信息科学版）》2014年第9期1103-1108,共6页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金资助项目(41171450;61001187);国家973计划资助项目(2011CB707100-6);中央高校基本科研业务费专项资金资助项目(6081002)~~

摘　　要：月面起伏较大导致嫦娥一号CCD多视影像之间的形变是非刚性的,传统的基于仿射变换配准的超分辨率重建方法很难取得理想的效果。本文提出了一种新的基于光流配准的超分辨率重建方法,该方法采用双边总变分(bilateral total variation,BTV)超分辨率重建方法迭代求解高分辨率影像,迭代初始值由光流配准后的多视影像共同产生,并在迭代求解高分辨率影像的过程中更新光流场。实验结果表明,本文方法解决了嫦娥一号CCD多视影像超分辨率重建中的亚像素精确配准问题,重建影像的空间分辨率显著提高,细节分辨能力增强。As the deformation among multi-view images from the Chang＇E-1 CCD is non-rigid, given the non-planarity of the moon surface, the traditional methods based on the affine transform model are not suitable for super-resolution correction. For this reason, a new optical flow based super-resolution framework is proposed. In the framework, the BTV method is employed to estimate a high resolution image iteratively. The initial values for a high resolution image are generated by all multi-view images that are registered by optical flow, while the optical flow fields are updated each iteration. Experimental results show that the proposed method performs better for the super-resolution of Chang＇E-1 CCD multi-view images than affine transform based methods. The resolution of the reconstructed image is greatly improved and more details are visible without ringing-effects.

关 键 词：嫦娥一号 CCD多视影像 超分辨率 光流 

分 类 号：P237.3[天文地球—摄影测量与遥感] TP751[天文地球—测绘科学与技术]