Clouds and Shadows Detection in Multi-spectral Satellite Image Based on Maximally Stable Extremal Regions

Maximally Stable Extremal Region (MSER) hasbeen proven to be an excellent feature extraction method incomputer vision. The initial MSER detector is merely appliedto intensity space. However, in some special cases whichcontain numerous data such as remote sensing, it is necessaryto acquire more information for more accurate analysis.Meanwhile, clouds and shadows detections of remote sensedimages are both fundamental and vital to accurateclassification and interpretation. Hence, this paper proposed anew method based on MSER to detect shadows and clouds inEnhanced Thematic Mapper Plus (ETM+) images, utilizingMSER in some particular spectrums with geographicalpractical backgrounds. In order to evaluate the effect ofMSER in different spectrum, several typical combinations ofbands were compared. Results indicated the favorable bandscombination for clouds detection, and advantageous bandscombination for shadow extraction respectively.