Cloud detection based on segmentation with statistical and geometry features

Cloud detection, recognition has been received increasing attention during last decades in remote sensing application field. We propose a novel cloud detection algorithm based on statistical region merging segmentation with statistical and geometry features. To distinguish clouds objects from a background, the statistical region merging segmentation algorithm is firstly adopted to obtain semantic segmentation regions. Based on information of segmented patches, statistical features, including spectrum and geometry features are extracted to represent otherness between clouds and underlying surface. Such features are finally implied to math the feature temple by the nearest neighbor algorithm. We show in this paper the addressed method make a effective cloud detection without any prior constraints and auxiliary data. Experiments have been carried out on aerial optical images to validate our proposed method.