风云气象卫星影像自动精细云检测

抠图技术能够准确细致地检测出前景与背景的差异，但是现有的大多数抠图算法均需事先给出前景与背景的标记，在海量影像处理业务中具有明显的局限性。本文将抠图技术引入风云气象卫星影像云检测中，提出了基于连通区域抠图的自动精细云检测算法。首先对影像的灰度图像进行多种阈值方法二值化，通过投票法集成得到初步二分图；在此基础上计算连通区域，以连通区域的重心为种子点，自动生成连通区域三分图；最后使用闭形式抠图法求解alpha值进而得到精细的云检测结果。将该方法与基于学习和稳健性抠图检测算法进行了对比，试验结果表明，该方法能够较好地检测出多种类型的云，具有更高的准确率。

Automatic and accurate cloud detection for Fengyun meteorological satellite

The matting technique can accurately detect the difference between the foreground and the background, but most of the existing matting algorithms need to give the foreground and background markings in advance, which has obvious limitations in the mass image processing. In this paper, the matting technology is introduced into the cloud detection of Fengyun meteorological satellite images, and an automatic fine cloud detection algorithm based on connected regions is proposed. Firstly, the gray visible image is binarized by ensemble threshold methods, and the preliminary bipartite map is obtained by the voting method. On this basis, the connected regions are calculated, and the center of gravity of the connected regions are taken as the seed point, and then triple map based on the connected regions is automatically generated. Finally, the closed-form matting algorithm is used to solve the alpha value to obtain fine cloud detection results. The proposed method is compared with the learning and robust matting-based cloud detection algorithms. The experiments show that the proposed method can automatically detect multiple types of clouds better and has higher accuracy.