图像质量评价在空间相机图像实时复原中的应用

	图像质量评价在空间相机图像实时复原中的应用
其他题名	Image Quality Assessment Applied to Real-time Image Restoration System on Space Camera
	王俊琦
学位类型	硕士
导师	张立国
	2015-10
学位授予单位	中国科学院大学
学位专业	光学工程
关键词	图像质量评价光学遥感图像复原信噪比调制传递函数清晰度对比度振铃
摘要	空间遥感相机的图像在轨实时复原已进入工程实施阶段，而在轨图像复原系统状态是否良好，经复原处理后的图像质量是否提高，需要选取相关的方法去评价。复原图像的质量能直接反应复原系统的性能及复原算法的有效性与稳定性，因此，本文开展了针对空间相机图像实时复原系统的图像质量评价，从图像的角度来间接评估系统性能。从遥感系统的评价及图像质量的评价两方面，选择了五项评价指标：信噪比与调制传递函数，清晰度、对比度及振铃波纹。论文对上述五项指标进行了详细研究，主要研究成果如下：在图像复原过程中，提高图像MTF与清晰度的同时，会在一定程度上放大噪声，噪声是衡量图像质量的重要参数之一。基于分块的方法设计了遥感图像噪声估计方法，将图像分为5*5的小块，计算每一个小块的方差，以小于阈值的子块作为平坦区域并标记出来，对所有标记区域的方差求平均，其结果作为图像的噪声估计。通过实验验证了该方法的精度以及在在轨信噪比估计中的有效性。调制传递函数MTF是光学成像系统的一个重要指标，它是反映空间频率传输性的函数。为了适应更多的遥感场景，在分析并对比已有的MTF估计方法后，提出了可以利用任意形状刃边图像计算MTF的方法，分析了算法步骤及图像选择等对结果产生的影响。运用MTFC技术，对不同图像进行复原，MTF值及图像的各项评价指标均得到提高，实验结果证明了任意形状刃边提取MTF的可行性和用于降质图像复原时的有效性。文章从全局与局部两个角度分别提出了两种清晰度的评价方法。从全局出发，基于骨架边缘提取的半参考清晰度评价，计算图像骨架边缘对应的像素数目，并以此作为清晰度评价参数，将降质图像与复原图像的清晰度参数作对比，得出复原后图像的清晰度提升率。通过对不同遥感图像的实验，验证了使用该方法进行图像清晰度评估的有效性与可靠性，同时证实了该方法具有抗噪声性能以及对弱振铃波纹的抑制效果。从局部入手，基于线扩散函数提取的无参考清晰度评价，截取图像中满足条件的目标刃边区域，对目标区域求梯度，选择梯度最大值点及其梯度方向作对应线扩散函数曲线，以线扩散函数曲线的半高全宽作为图像的清晰度指标。分别从对同源图像的清晰度评价，算法抗噪声性能，对不同图像的清晰度评价三个方面进行实验，结果表明，该方法能够评价图像的清晰度，不需要参考图像且具有良好的抗噪声性能。在遥感系统成像以及图像复原后，图像的对比度均会发生变化。总结并对比已有的对比度评价方法，选择Weber对比度作为本文的对比度评价指标。频域的高频截断使复原图像出现了振铃，PSF估计越不准确，振铃越大。获取纹理简单区域的边缘及边缘附近振铃，以振铃与边缘之间的比值作为图像的振铃评价值。为了方便使用，设计了空间相机在轨复原图像质量评价工具箱。工具箱具有简单的图像预处理功能，能去除图像条带噪声，粗略估算原始图像对应的点扩散函数并对其复原。工具箱读取复原前后的图像，分别应用不同的指标进行评价，各项指标的评价结果显示在对应界面上。
其他摘要	On-orbit image restoration of the space remote-sensing camera have been realized, however, whether the on-orbit image restoration system functions well, and whether the image quality have been elevated after recovery, it needs some methods to estimate. The quality of the restored images can directly reflect the performance of the recovery system and the validity and stability of recovery algorithms, therefore, this paper has carried out the image quality assessment for on-orbit image restoration system, and evaluate the system capability from image. Five evaluate parameters are chosen based on remote sensing system metrics and image quality assessments, which are signal-to-noise ratio, modulation transfer function, sharpness, contrast and ringing. The main research results are as follows: In the process of image restoration, it will be in a certain extent to enlarge the noise when improving the image MTF and the clarity, noise is an important parameter on image quality assessment. Estimating the noise of remote sensing image is designed based on the method of block. The image is divided into small pieces of 5*5, calculate the variance of each block, mark these blocks which variance lower than the global variance, count the average of marked blocks, The results are obtained as the image noise estimation. The accuracy and the validity of the proposed method are verified by experiments. Modulation transfer function is an important index of optical imaging system, which is a function of spatial frequency transmission. After analyzing and comparing the slanted-edge method, a new method is proposed to calculate the MTF with arbitrary shape edge image, analyzed the effect of the algorithm steps and image selection. Using the MTFC technology to reconstruct different images, the MTF value and the evaluation indexes are improved, the experimental results prove the feasibility of extracting MTF with arbitrary shape edge image, and the effectiveness of image restoration. In this paper, two kinds of clarity assessments are proposed. From the global perspective, extract the image skeleton, use the number of skeleton pixels as sharpness parameter, compare the sharpness parameters between distorted image and recovered image, calculate the sharpness upgrade rate. Through the experiments of different remote sensing images to verify the validity and reliability of image sharpness evaluation using the method, also confirmed that the method has anti-noise performance and the weak ringing ripples inhibitory effect. From the local start, the no-reference resolution based on extracting the line spread function of target edge region in arbitrary image. Compute the gradient of target region, find the maximum gradient point and its direction to draw the line spread function curve, use the full width at half maximum as the image sharpness parameter. From the definition of the homologous image, the performance of the algorithm is tested. The results show that the method can evaluate the image's clarity, without the need of reference image and has good anti-noise performance. In the remote sensing system, the illumination is insufficient, and the dynamic range of the imaging sensor is too small will lead to low contrast. At the same time, the contrast of the image will be changed after image restoration. Summarize and compare the current contrast evaluation method, select the Weber contrast as the contrast evaluation index of this paper. Due to the high frequency cut-off in frequency domain make image appeared ringing, the less accurate of PSF estimation, the greater ringing. To obtain simple texture region edge and fringe ringing, use the the ratio between the ringing and the fringe as ringing metric. An image quality assessment toolbox is designed to facilitate the use. The toolbox has simple image preprocessing functions, it can wipe off the push-broom noise, as well as the original image corresponding to the point spread function and its recovery. Read the image before and after restoration, apply the different indexes to evaluate. The assessment results are shown in the corresponding interface.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/49324
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	王俊琦. 图像质量评价在空间相机图像实时复原中的应用[D]. 中国科学院大学,2015.

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
王俊琦.pdf（2556KB）	学位论文		开放获取	CC BY-NC-SA	请求全文