导　　师： 杨建峰;薛彬

学科专业： 0803

授予学位： 硕士

作　　者： ;

机构地区： 中国科学院研究生院

摘　　要： 嫦娥一号（CE-1）卫星CCD立体相机完成了对全月表立体成像，并获得了高像质的月表影像数据。为了进一步提高影像的空间分辨率，从而对CE-1月表影像数据进行深层次地挖掘，需要开展月表影像的超分辨率技术研究。同时,该研究也将为CE-2卫星影像超分辨率实现奠定基础。本文的主要内容包括：CE-1卫星CCD鉴定级相机点扩散函数测量、月表影像超分辨率复原算法实现和超分辨率结果评价等。 本文首先详细分析了CE-1卫星月表影像的数据特点，并结合超分辨率的理论基础和算法给出月表影像的超分辨率复原模型。 其次介绍了超分辨率先验参数的测量，其中主要完成了CCD相机点扩散函数的测量实验，包括实验原理、实验设计与实现以及最终的实验数据后处理等。其中实验原理的核心思想是采用亚像元位移推扫方式完成对像面的横向上（过）采样，然后通过拟合手段近似求得二维点扩散函数。最后考虑到卫星与月表的相对运动，对实测的点扩散函数引入了运动模糊，得出最终的点扩散函数。 最后，对于超分辨率复原算法实现部分，本文选用最大后验概率法（MAP）进行月表影像超分辨率复原。其中详细介绍了目标函数的实例化模型和目标函数求解方法。实例化模型采用Gauss模型和Gibbs模型，目标函数的求解采用共轭梯度法。然后采用模拟线对实验测试算法的复原能力及适用范围，结果表明，该算法可用于月表影像的超分辨率复原，并且可实现分辨率提高至1.36倍以上的目的。最后给出了超分辨率复原结果，并且从主观视觉、客观指标（信息熵和清晰度）以及频谱等方面对复原结果进行了评价，结果表明复原图像像质优良。 Chang e1/(CE-1/) satellite CCD stereo camera has completed athree-dimensional imaging on the moon, and obtained great images of the lunarsurface. In order to further improve the spatial resolution of the lunar images toachieve excavating the CE-1lunar image data, we need to carry out super-resolutionresearch of the images. Meanwhile, this research will also lay the foundation for CE-2satellite super-resolution recovery. The main contents of this paper include: themeasurement of CE-1satellite CCD camera point spread function, thesuper-resolution algorithm for lunar surface images, and the evaluation of thesuper-resolution results. To begin with, after a detailed analysis of the CE-1satellite lunar datacharacteristics, combined with the super-resolution theoretical foundation andalgorithm, the super-resolution recovery model applied to the lunar images isobtained. Then, we introduce the measurement process of the super-resolution prioriparameters in depth, in which we mainly complete measuring point spread function ofCCD camera, including the purpose of the experiment, experimental principle, designand implementation and experimental data processing. Among them, the core idea ofexperimental principle is over-sampling the image plane by the way of sub-pixeldisplacement, and then, we get the two-dimensional point spread function by fitting.Finally, taking the relative motion between the satellite and the lunar surface intoaccount, we modify the obtained point spread function with the motion blur, and getthe final point spread function. In the end, we select the Maximum A Posteriori method /(MAP/) as thesuper-resolution algorithm of the recovery. Then, we introduce the instantiated modeland the solving method of the objective function in detail. Among them, the Gaussmodel and the Gibbs model are chose to instantiate the objective function and theconjugate gradient method is used to solve it. Next, simulated pairs are used to test theresilience and the application of the algorithm, and the results show that this algorithmcan be used for super-resolution recovery of the lunar images and can improve theresolution of more than1.36times. Finally, we show the super-resolution results, andevaluate them in terms of the subjective visual, objective indicators /(entropy anddefinition/), and the spectrum respectively, which show that the quality of thesuper-resolution images is great.

关 键 词： 嫦娥一号月表影像 超分辨率 点扩散函数 最大后验概率法 共轭梯度法

分 类 号： [TP391.41]

领　　域： [自动化与计算机技术] [自动化与计算机技术]