光学遥感影像压缩及融合的质量评价研究

发布时间：2019-06-12 05:23

【摘要】：由于遥感卫星成像系统性能、图像压缩算法、数据远程传输设备等的不完善,以及大气层及设备噪声等的干扰,在遥感卫星图像的获取、压缩、传送的过程中,遥感图像降质和模糊失真是不可避免的。这些给遥感图像的处理、分析、应用带来了很大的难度。本文针对遥感图像压缩算法对比选优中,对压缩图像的主观评价及客观评价的实际问题,以及目前国标遥感影像图质量评价中,只有主观定性描述,没有客观量化指标的问题,在广泛了解国内外图像质量评价相关理论和技术的最新发展现状基础之上,分析和研究了遥感图像质量的主观评价方法和客观评价方法存在的问题,并探讨了原始遥感影像及遥感融合影像的质量评价的方法和思路,为进一步的深入研究奠定了基础。本文的主要研究工作如下：一、图像质量主观评价方法研究针对遥感图像压缩算法上星选优的实际问题,提出了一种新的基于Shell排序算法的图像质量主观评价方法,并开发了基于该方法的遥感图像主观评价软件。该方法是将Shell排序算法与已有的遥感压缩图像质量的主观评价方法中的成对比较法相结合,兼有排序方法和成对比较法的优点,减少了主观评价中图像比较的次数。利用该软件,对924幅遥感压缩图像,经过两组遥感专业一线作业人员的主观测评实验,最后对实验结果进行分析处理,结果表明新主观评价方法比原有的成对比较法提高效率达38%-52%。二、图像质量客观评价方法研究为了从客观的角度评价遥感压缩失真图像的质量,本文提出一个将图像的梯度幅值、相位以及结构相似度(SSIM)三者相结合的图像质量评价新模型—梯度相似度(GSIM)模型,以及基于该模型的图像质量评价算法。新模型与SSIM模型及基于梯度的模型相比,不仅包含亮度、对比度和结构三部分信息,而且更重要的是该模型增加了梯度相位信息。通过对LIVE图像数据库的982幅失真图像和924幅遥感压缩影像的实验,结果显示新模型的性能优于MSE、PSNR、SSIM等传统模型以及基于梯度的模型。与SSIM等模型相比,新模型不但能较好地解决对严重失真图像的客观评价与主观感受并不完全相符的问题,而且还能更好地处理对多种类型失真图像的混合评价效果较差的问题。三、遥感影像产品的构象质量评价研究针对遥感影像平面图制作规范国标(GBT15968-2008)中,对影像质量评定标准,只有主观定性描述(“层次丰富、清晰易读、色调均匀,反差适中”),而无客观定量的评价指标的问题。对于图像的层次、清晰度、辐射、反差、信噪比等方面的表达和评价问题,本文从亮度、对比度、信息量、清晰度、纹理信息及空间细节的角度进行分析,分别给出相应的客观评价指标。对于图像对比度反差适中的主观定性描述,在深入研究人眼视觉特性的基础上,提出了一种的基于高斯正态函数加权的对比度评价指标,用该指标表示反差适中,比标准差更符合人的视觉感受,最后用实验验证新模型指标的有效性。四、遥感影像融合及其质量评价研究本文针对遥感影像生产单位实际对遥感影像融合算法的选择问题,详细比较了现有的10种遥感影像融合算法,研究了各种融合算法的基本原理,通过各种图像质量评价指标,分析了各种融合算法的特点,最后提出了一般遥感影像构象质量的评价体系,将遥感影像的构象质量分解为亮度、对比度、清晰度、信息量、光谱信息、纹理信息、信噪比等几方面因素来分别考虑,并在现有客观指标深入研究的基础上,对上述各方面都给出了具体的客观评价指标。
[Abstract]:In the process of acquisition, compression and transmission of remote sensing satellite images, remote sensing image degradation and fuzzy distortion are inevitable due to the imperfections of remote sensing satellite imaging system performance, image compression algorithm, data remote transmission equipment and the like, as well as the interference of the atmosphere and equipment noise. The processing, analysis and application of these remote sensing images bring great difficulty. In this paper, based on the comparison of remote sensing image compression algorithms, the actual problems of the subjective evaluation and objective evaluation of the compressed image, as well as the current national standard remote sensing image quality evaluation, are only the subjective description, and there is no objective quantitative index. On the basis of a wide understanding of the latest development of the relevant theories and techniques of image quality evaluation at home and abroad, the paper analyzes and studies the subjective evaluation method and objective evaluation method of the remote sensing image quality. In this paper, the method and thought of the quality evaluation of the original remote sensing image and the remote sensing fusion image are discussed, and the basis of further research is laid. The main research work of this paper is as follows: Next: subjective evaluation of image quality In this paper, a new image quality master based on shell sort algorithm is proposed in order to solve the practical problem of the satellite selection on the remote sensing image compression algorithm. The method of view evaluation is developed and a remote sensing image master based on the method is developed. The method is combined with the paired comparison method in the subjective evaluation method of the quality of the existing remote sensing compressed image, and has the advantages of the sorting method and the comparison method, so that the image in the subjective evaluation is reduced, The results show that the new subjective evaluation method is more efficient than that of the former one, and the result shows that the new subjective evaluation method is more efficient than the original one. % -52%.2, Image quality In order to evaluate the quality of the remote sensing compression distortion image from an objective point of view, an objective evaluation method is presented in this paper. a new model of image quality evaluation based on the combination of gradient magnitude, phase, and structure similarity (SSIM) of the image is used to evaluate the gradient similarity (GSIM) model of the new model, and based on the model, Compared with the SSIM model and the gradient-based model, the new model not only contains three parts of the brightness, the contrast and the structure, but also the model The results show that the performance of the new model is better than that of MSE, PSNR, SSIM and other traditional models. Compared with the SSIM and other models, the new model can not only solve the problem that the objective evaluation of the serious distorted image is not completely consistent with the subjective feeling, but also can better deal with the mixing of many types of distorted images. the problem of poor evaluation.3. Remote sensing image In the national standard (GBT15968-2008), the study on the conformation quality of the product has only a subjective description ("The level is rich, clear and easy to read, the tone is uniform, the contrast is moderate") for the standard of image quality assessment (GBT15968-2008). There is no objective and quantitative evaluation index. For the expression and evaluation of the level, definition, radiation, contrast and signal-to-noise ratio of the image, this paper analyzes the angle of brightness, contrast, information amount, definition, texture information and spatial detail. The objective evaluation index is given respectively. Based on the study of the visual characteristics of the human eye, a kind of contrast evaluation index based on the Gaussian positive state function weighting is proposed based on the study of the visual characteristics of the human eye. The contrast is moderate. The specific standard deviation is more in line with the human visual sense, and finally, The experiment verifies the effectiveness of the new model index. In this paper, the remote sensing image fusion and its quality evaluation are studied in this paper. In this paper, the selection of the remote sensing image fusion algorithm for remote sensing image production units is studied in this paper, and the existing 10 remote sensing image fusion is compared in detail. In this paper, the basic principle of various fusion algorithms is studied, the characteristics of various fusion algorithms are analyzed through various image quality evaluation indexes, and finally the evaluation system of the constellation quality of the general remote sensing image is put forward, and the conformation quality of the remote sensing image is decomposed into the brightness and the contrast, The definition, information amount, spectral information, texture information, signal-to-noise ratio and the like are taken into consideration respectively, and on the basis of the in-depth study of the existing objective indexes, the invention
【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】