【作者】 江兴方；

【导师】 陶纯堪；

【作者基本信息】 南京理工大学 ， 光学工程， 2007， 博士

【摘要】 随着光学图像增强方法研究的不断深入,特别是遥感数据的应用日益广泛,对制约遥感图像质量的重要因素—云的研究显得越来越重要。由于存在云噪声,大多数遥感数据不能获得应有的信息,因此在遥感图像中获取地物的信息,必须研究去云的方法。常用的去云方法有多光谱图像法、多幅图像插值法、多源数据融合法、同态滤波法以及采用对云不敏感的传感器方法等等,在一定的条件,对灰度图像较为有效。对于无先验知识的单幅有云噪声的遥感图像,则需要具体问题作具体分析。对光学图像去云方法的研究,首先建立了云噪声影像模型,分析了遥感图像中云的特征,基于波的传播原理,提出了用惠更斯次波法提取遥感图像中云区和云影区,采用边缘模糊的方法,增强所提取区域的信息,进而实现遥感图像局部去云和去云影:基于云区渐变的特性,提出了用高次多项式曲线变换法,拉伸云区像素的灰度值,突出了云区图像的信息。其次基于邻域法分析了同态滤波法的本质,就是在图像中滤去低频信号,正因为云在遥感图像中正好属于低频信息,故同态滤波法也能用于遥感图像去云。深入研究了E.Land在20世纪60年代提出的Retinex思想,以及无数学者在Land的基础上研究获得的各种各样算法,特别研究了工作在NASA和CWM的D.J.Jobson,Zia-ur Rahman,G.A.Woodell(简称JRW)等人的Retinex思想。揭示了Retinex算法的物理本质,提出了改进型多尺度Retinex算法:运用多尺度Retinex算法增强彩色图像后,在其亮度平均值附近以k倍标准差截取,再进行拉伸,在得到的新图像中亮度与对比度的乘积落在[6000,10000]范围内的图像,质量最佳。并得到结论:改进型多尺度Retinex算法增强曝光不足的彩色图像,以k=1倍标准差截取后再拉伸得到的图像最好。改进型多尺度Retinex算法应用于彩色遥感图像去云中,方法是先进行取补色,再应用改进型多尺度Retinex算法,突出云区域中的信息,最后再取补色,这一新方法去除薄云效果最佳。最后,遥感图像去云方法推广到除去高大建筑物的阴影上。采用边缘提取的方法,改进了闭合区域提取的软件,提出了真断点、准断点的判据,将邻点的邻点命名为特征点,有效地提取图像中的闭合区域,将Retinex增强后的图像相应的区域替代后,复合的图像突出了阴影区域的信息。基于小波变换,尝试了基于小波变换的图像增强方法,利用DSP采用小波变换的算法,用CCS编程语言编写边缘提取的工程文件,采用比较法,在显示屏上一半为所摄图像,另一半为实时提取图像中的边缘的图像,效果好。为研究基于DSP实时去雾的方法做好了准备。更多还原

【Abstract】 The research work of cloud that makes the image quality lower is more important with the development of the optical images application, in particular, of the remote sensing data application. We can not capture the information from the most remote sensing data because there is cloud. So the methods of removing cloud from the remote sensing image must be researched. There are several methods for removing cloud in some conditions and they suit for back and white remote sensing images, such as the methods of Multi-spectrum image, Multi-image interpolation, Multi-source data fusing, and the method of Homomorphic filtering. The concrete methods of removing cloud for a single remote sensing image without any pre-information need research.The research of this paper consisted of three aspects, which are described of image elements, neighborhood regions and wavelet transform.First, the image model of the cloud noise had been build, and the sharp characteristics of the cloud noise had been analyzed. The paper suggested the new method for extracting cloud region and its shadow region. The new method is the method of Huyghens secondary. The local information of the remote sensing images in cloud and its shadow could be protruded by the new method. Considered the gradation various characteristics of the cloud, another method that was the method of higher-degree polynomial curve transform had been suggested. The hidden information in cloud region was exposed by the method.Second, the original quality of Homomorphic filtering method that removed lower frequency signal of original images was analyzed. The cloud in the remote sensing image is just the lower frequency signal of original images. It was researched that the Retinex theory proposed by E. Land on December 30, 1963. The algorithms of Retinex were studied by this paper. In particular, the Center / Surround algorithm that proposed by D. J. Jobson, Zia-ur Rahman, G. A. Woodell (JRW) was considered. The physics original quality of Retinex had been revealed. The advanced multi scale Retinex (AMSR) algorithm was put forward: After the image enhanced by multi scale Retinex the truncate interval of [μ- kσ,μ+ kσ] near the mean brightnessμin k times standard deviationσrange had been elected and stretched. If the production of the brightness multiply with the contrast degree for the stretched image was in [6 000, 10 000], the stretched image has high quality. As the same time, the conclusion has been proved: The method of highest quality for the underexposure color image was that it selected k=1 time standard deviation range near the mean brightness and stretched after enhanced by multi scale Retinex.In the process for the application of advanced multi scale Retinex algorithm in remote sensing images, there are three steps. The first step elected the complementary color image of the original image. The second step applied the method of the advanced multi scale Retinex algorithm to enhance the information in dark region that was just the cloud region. The last step elected the complementary color image of the enhanced image by the advanced multi scale Retinex. The results of experiments shown the method was effective.Finally, two of the application fields of the removing cloud methods had practiced. The cloud regions had been extended to the shadow of the grant builds under the sun. The method of removing cloud had been extended to remove fog, and so on. In the first application, this paper extracted the edge of pyramids in Egypt based on Canny algorithm and programs. The programs were in a software package that developed by our theme group was used to extract the shadow regions in remote sensing images. The shadow regions were placed by the same regions that enhanced by multi scale Retinex algorithm and the more information in the shadow was showed. The software package was improved. The characteristic point idea was proposed and the distinguish method between the real-broken point and quasi-broken point was described. The closed regions in the remote sensing images were extracted effectively. Another application was that the edge details were extracted in real time with wavelet transform based on DSP. The results of extracting edge details that used the project software programmed by CCS were good. The application was general preparing for the issues of removing fog in real time based on DSR更多还原