摘要
对云及其阴影的识别是遥感图像处理中的一项基础性工作,在高分辨率遥感影像中,云及其阴影在图像中的分布是有规律的,利用两者在平坦区域高分辨率卫星影像上具有相似性的特征对其进行识别与匹配,可以比较简单地利用图像域值分割方法得到更好的识别与匹配结果。采用面向对象的思路提取云及其阴影的轮廓,在分析图像分割误差原因的基础上,考虑影像上云与其阴影的空间拓扑关系,应用改进的分数Hausdorff距离的图像匹配方法(MPHD),通过云及其阴影的局部相似的匹配,从而很好地识别出云及与其匹配的阴影,同时还可计算出匹配两者在投影平面上的距离。提出的云及其阴影的识别与匹配算法,为计算云高和应用遥感图像处理云及其云阴影的掩模提供科学依据。
Clouds are the obstruction of visual and infrared remote sensing and their shadows may also lead to an intolerable bias of the true reflection of the underlying elements. Thus a reliable cloud and shadow mask is essential for further processing. Clouds cast shadows on the earth＇s surface. In high resolution remote sensing images, it is common that different objects have the same spectral characteristics, so clouds are always confused with objects such as concrete surface, house roof and bare land, and shadows are always confused with water when extracting them based on spectral. Clouds＇ profiles. So a robust image matching algorithm called Modified Partial Hausforff Distance（MPHD） is introduced to find out the match of every cloud with its shadow and then calculate the pixel distance between them. Firstly, threshold segmentation is carried out to divide the image into clouds, shadows and backgrounds and to get the edge of the cloud and shadows. Then topology analysis is made to exclude edges that caused by over segmentation and coverage between clouds and shadows and the algorithm called Modified Partial Hausdorff Distance （MPHD） is used to match the clouds and shadows. This method is high in the stability of the covered edge and can calculate the positional offsets between matched clouds and their shadows. Before the matching, prior knowledge is added to make the match of every cloud with the shadow along the direction of sunshine. To reduce the computational complexity, MPHD was modified. After matching operation match pairs, match values and positional data are obtained. The match value and match pairs are of the characteristics of direct proportion. Based on the analysis of the relationship between match value and positional data, the most accurate offset at the minimum match value （when the two match the best） was obtained. Tests were made with some SPOT 5 remote sensing images, in which the exposed concrete ground of the dock with highest grey value was recognized as clouds and the cultivati
出处
《海洋学研究》
2009年第2期51-57,共7页
Journal of Marine Sciences
