基于同场景多源数据先验信息的遥感图像半盲恢复研究

项目名称： 基于同场景多源数据先验信息的遥感图像半盲恢复研究

项目编号： No.41471368

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 天文学、地球科学

项目作者： 刘鹏

作者单位： 中国科学院遥感与数字地球研究所

项目金额： 70万元

中文摘要： 遥感图像的盲恢复需要同时估计降晰函数和原始图像，由于未知变量过多而且问题严重病态导致求解非常困难。本项目提出利用同场景多源遥感数据作为先验知识辅助进行未知降晰函数条件下的遥感图像恢复（此时称为半盲恢复）。研究内容主要为：基于多源参考图像统计特性优化降晰函数的求解；联合利用参考图像与目标图像（也称为估计图像）的特征改善图像恢复效果；参考图像的噪声、配准误差和纹理差异对图像半盲恢复的影响。本项目的特色是有效利用遥感领域日益增多的数据源，将同场景多源遥感数据作为先验信息约束图像盲恢复的计算过程，其创新之处主要在于：一方面，建立了参考图像、目标图像与降晰函数求解三者之间相互依赖的关系；另一方面，基于参考图像与目标图像的纹理关系构造新的规整化项，有利于抑制噪声并更多地恢复出图像的细节。拟解决的科学问题是：如何有效利用参考图像中有利的先验信息和如何同时规避参考图像所附带不利因素的影响。

中文关键词： 数据融合；图像恢复

英文摘要： Blind remote sensing image restoration is very difficult, which is a serious ill-posed problem, because it needs to solve both the original image and degradation kernel at the same time. In this proposal, we propose to use multi-source remote sensing data as reference priors to aid the image restoration without knowing the blur kernel (we called it as semi-blind image restoration). The major research contents are: estimate the blur kernel based on the statistical characters of the reference image; combine the features of observed image and the reference images to suppress the noise in de-convolution and recover more details; the influence from the from noise of reference images, image registration errors and texture differences. The novelty of the proposal is: by effectively utilizing the multi-source remote sensing data, we introduce the different priors' information into the blind image restoration to optimize the computation. One hand, We constructed the relationship between reference images, object images and the estimation of blur kernel; the other hand, we proposed the construct a new hybrid regularization term to better suppress noise and recovery more image details. The scientific problem to be solved is: how to employ the useful prior information in the reference image and at the same time how to avoid the negative influence from the reference images.

英文关键词： data fusion;image restoration