航空图像超分辨率重构技术研究

	航空图像超分辨率重构技术研究
	杨文波
学位类型	博士
导师	朱明
	2014-07
学位授予单位	中国科学院大学
学位专业	机械电子工程
摘要	随着航空技术日新月异的发展，航空成像技术已经在地形测绘、土地和森林资源调查、城市建设、铁路和公路建设以及军事侦察等诸多领域得到了广泛的应用。但在航空摄影的过程中，图像的质量受诸多因素的影响。由于受太阳光散射引起的空中雾蒙的亮度、航空摄影时飞机发动机的震动、曝光时间内飞机运动引起的像点移位、成像系统物镜的残余像差、感光材料的性能以及飞行姿态等因素影响，使实际获得图像的分辨率较低，图像质量较差。因此，研究航空图像的超分辨率重构，从图像处理的角度提高航空图像的分辨能力，具有重要的应用意义。本文着重研究航空图像超分辨率重构的一些关键技术问题，研究成果如下： (1) 单帧图像超分辨率重构。针对同一场景不具备拍摄多帧图像的条件，本文重点研究了线性插值、3次Ｂ样条插值、O-MOMS插值和Keys插值四种多项式插值重构算法，给出了四种多项式线性空间移不变的插值公式，但是线性移不变插值技术易导致图像边缘平滑，为了解决这一问题，本文提出了一种线性空间移变自适应插值算法，该算法是对有偏距离插值算法的改进，对多项式插值中的像素值赋以不同权值，像素值的权值因子大小取决于其邻域像素值不一致性，权值因子能使图像局部均匀一致，有助于保持图像局部连续性，提高重构图像的质量。 (2) 序列图像超分辨率重构。本文将互信息配准技术应用到航空图像序列的配准中，用来提高图像配准的精度与鲁棒性。针对航空图像大数据量的特点，本文提出了一种基于先验信息的超分辨率重构算法。该算法把图像复原和信息融合分开处理来提高运算效率，利用比较简单的非迭代算法进行图像复原，再用小波变换的方法将图像序列中所包含的冗余信息和互补信息融合到一帧图像中，最后用多项式插值重构高分辨率图像。该算法的核心是去模糊，由于模糊矩阵是分块Toeplitz循环阵，因此存在非奇异矩阵使其对角化，避免了对大型稀疏矩阵直接求逆，加快了运算速度。本文分别给出了线性最小均方误差法、最大信息熵法、正则化方法三种去模糊的非迭代数值算法。针对航空图像模糊核有时难以进行有效的估计，导致观测模型无法精确的建立，本文提出了一种基于最大公因子法的盲超分辨率重构算法。该算法把图像复原和信息融合分开处理，是以模糊核的盲估计为基础的。对于图像的盲复原，本文采用二维最大公因子盲复原算法，由于二维最大公因子盲复原算法仅针对两帧图像进行恢复，因此，本文通过序列图像的每一帧信息，重新生成一帧观测图像，给出了序列图像的二维最大公因子盲复原算法。 (3)亚像元成像增加探测器的时间和空间采样频率，可以提高亚像元成像系统空间分辨率。但是探测器采集到的数据有混叠，重构得到的高分辨率图像发生模糊，分辨率远未达到理想值，为了解决这一问题。本文提出一种针对多线阵亚像元成像的超分辨率重构算法。首先，在高分辨率网格上建立插值模型；然后，辨识插值重构图像在线阵列方向和扫描方向的模糊核，得到整幅图像的模糊核；最后，采用带有Neumman边界条件的梯度平滑正则化模型去除模糊。
其他摘要	With the rapid development of aviation technology, aviation imaging technology has been widely applied in such fields as topographic mapping, land and forest resources survey, urban construction, railway and highway construction, military reconnaissance, and so on. But in the course of aerial photography, the image quality depends on many factors. Influenced by such factors as brightness of mask fog in air caused by the scattering of sunlight, vibration of aircraft engine in course of aerial photography, displacement of image point caused by aircraft movement within exposure time, the residual aberration of the objective lens of the imaging system, the performance of the photosensitive material, the flight posture and so on, the virtually obtained image is inferior in both resolution and image quality. Therefore, it is of great significance in application to study super-resolution reconstruction of aviation image aiming to improve its resolution performance from the perspective of image processing. This paper puts emphasis on some of the crucial technical issues on super-resolution reconstruction of aviation image, the research results of which are as follows: (1) Single-frame image super-Resolution reconstruction. In view of the multi-frame image capture being not available at the same scene, this paper, based on the four polynomial interpolation algorithms for image reconstruction :linear interpolation, cubic B-spline interpolation, O-MOMS interpolation and Keys interpolation, introduces four linear space-invariant polynomial c interpolation formula. The pixel values of polynomial interpolation are endowed with different weights, and the weight factor of the pixel value depends on the asymmetry in its neighboring pixel values. Weight factor makes the image locally homogeneous, contributes to the continuity of the image locally, and improves the quality of the reconstructed image. (2) Sequence image super-resolution reconstruction. This paper applies mutual information registration techniques to image registration aviation sequences to improve the accuracy and robustness of image registration. Responding to large amounts of data in aviation images, this paper presents a highly efficient super-resolution reconstruction algorithm based on wavelet fusion. The algorithm processes the image restoration and information fusion separately to improve computational efficiency. It uses relatively simple non-iterative algorithm to restore image, and then fuse of redundant information contained in the image sequence and complementary information into one-frame image by wavelet transformation, and last reconstructs high-resolution image using polynomial interpolation. The core of the algorithm is debluring. The degradation matrix is a block Toeplitz-To-block circulant matrix and the diagonalization property via non-singular matrix in order to avoid directly inversing large sparse matrix and accelerate computational speed. This paper proposes three non-iterative algorithms for image deblurring:a linear minimum mean square error, the maximum entropy , regularization method. In dealing with the problem of the estimation of blur kernel in aviation images, which leads to failure to establish a precise observation model, this paper presents algorithm of blind super-resolution reconstruction based on wavelet fusion. The algorithm，based on blind blur kernel estimation, deal with the image restoration and information fusion separately. Because of the 2D-GCD algorithm is only to benefit from two observations rather than all observations, we suggest generating a new observation image incorporated the information from all the observations into the restoration , and the algorithm of blind restoration of images applying the greatest two-dimensional common divisor is presenting by this paper. (3) Spatial resolution of the sub-pixel imaging system can be improved by an increase of temporal and spatial sampling frequency in detectors. However, there exists aliasing in data collected by detectors and blurring in an image with high-resolution, as a result of which the resolution is far away from the ideal value. To resolve the problem, an algorithm of super-resolution reconstruction concerning multi-linear array sub-pixel imaging is put forward in the paper. First of all, an interpolation model on high-resolution grid is established. Next, blur kernels in an image with high-resolution are identified in linear array and scanning direction respectively, from which the blur kernel in a frame is obtained. Last, a model of gradient smoothing regularization with Neumman boundary conditions is employed to deblur.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/41493
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	杨文波. 航空图像超分辨率重构技术研究[D]. 中国科学院大学,2014.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
杨文波.pdf（3074KB）			开放获取	CC BY-NC-SA	请求全文