Clouds and their shadow removal in remote sensing image

Taiwan is an island located at monsoon climate section in the Pacific Ocean. There are often clouds in the sky which poses restrictions for the development of optical remote sensing. Cloudless images over the island will take months to retrieve. For the sensors of visible and infrared light on earth resource satellites, the interference of cloud and haze becomes a serious issue to deal with. In this case, we have to locate the cloud and haze from the satellite image. We first convert the image from (Red, Green, Blue) RGB system to the (Hue, Saturate, Intensity) HSI system to enhance the clouds and haze. Then a supervised pure pixel classification with minimum Euclidean distance is adopted for endmember classification. Six types of endmembers are selected and they are cloud, haze, vegetation, building, shadow and sea water. However, the cloud shadow should be separated from the terrain shadow. We expect the correlation coefficient can provide useful information, but this is still under investigation.