测绘学报 ›› 2016, Vol. 45 ›› Issue (9): 1081-1088.doi: 10.11947/j.AGCS.2016.20150451

• 摄影测量学与遥感 • 上一篇    下一篇

多尺度细节增强的遥感影像超分辨率重建

朱红1,2, 宋伟东1, 谭海2, 王竞雪1   

  1. 1. 辽宁工程技术大学测绘与地理科学学院, 辽宁 阜新 123000;
    2. 国家测绘地理信息局卫星测绘应用中心, 北京 100048
  • 收稿日期:2015-09-06 修回日期:2016-03-20 出版日期:2016-09-20 发布日期:2016-09-29
  • 通讯作者: 宋伟东 E-mail:lntu_swd@163.com
  • 作者简介:朱红(1989-),女,博士生,研究方向为遥感影像超分辨率重建、影像匹配等.E-mail:zhuhong19890408@163.com
  • 基金资助:
    测绘地理信息公益性行业科研专项(201412007);辽宁省科技博士启动基金(20141142)

Remote Sensing Images Super Resolution Reconstruction Based on Multi-scale Detail Enhancement

ZHU Hong1,2, SONG Weidong1, TAN Hai2, WANG Jingxue1   

  1. 1. School of Geomatics, Liaoning Technical University, Fuxin 123000, China;
    2. Satellite Surveying and Mapping Application Center of NASG, Beijing 100048, China
  • Received:2015-09-06 Revised:2016-03-20 Online:2016-09-20 Published:2016-09-29
  • Supported by:
    The Public Welfare Industry Scientific Research Based on Surveying and Mapping Geographic Information(NO.201412007);The Natural Science Foundation of Dr. Initial Funding of Liaoning Province(No.20141142)

摘要: 鉴于现有超分辨率重建方法难以突显重建影像细节信息的问题,提出多尺度细节增强的遥感影像超分辨率重建模型框架。首先,通过最小二乘滤波方法将序列影像分解成包含大尺度边缘的平滑信息和包含中小型尺度的细节信息;其次,利用插值方法得到相应的高分辨率细节信息和平滑信息,构造纹理细节增强函数,提升中小型细节的增强幅度;最终,融合细节信息和平滑信息,得到初始的超分辨率重建结果,并利用局部优化模型进一步改善重建影像质量。选取同时相和多时相遥感影像作为试验数据。试验结果表明,本文重建结果与插值方法、TV方法和MAP方法相比,在客观评价指标上均有显著提高,明显改善了重建影像的纹理细节。论文提出的多尺度细节增强的超分辨率重建方法,可以使重建影像提供更多高频细节信息,具有较好鲁棒性和普适性。

关键词: 超分辨率重建, 细节增强, 多尺度分解, 最小二乘滤波

Abstract: The existing methods are hard to highlight the details after super resolution reconstruction, so it is proposed a super-resolution model frame to enhance the multi-scale details. Firstly, the sequence images are multi-scale deposed to keep the edge structure and the deposed multi-scale image information are differenced. Then, the smoothing information and detail information are interpolated, and a texture detail enhancement function is built to improve the scope of small details. Finally, the coarse-scale image information and small-medium-scale information are confused to get the premier super-resolution reconstruction result, and a local optimizing model is built to further promote the premier image quality. The experiments on the same period and different period remote sensing images show that the objective evaluation index are both largely improved comparing with the interpolation method, traditional total variation(TV)method,and maximum a posterior(MAP) method. The details of the reconstruction image are improved distinctly. The reconstruction image produced using the proposed method provides more high frequency details, and the method proves to be robust and universal for different kinds of satellite remote sensing images.

Key words: super resolution reconstruction, detail enhancement, multi-scale decomposition,