Automatic Thick Cloud Removal for MODIS Remote Sensing Imagery

Abstract

In this paper, an automatic method is proposed to realize thick cloud removal for moderate resolution imaging spectrum-radiometer (MODIS) remote sensing imagery. The proposed method can make full use of MODIS advantages of high temporal resolution and spatial resolution. The overlapping region can be detected by utilizing geographical information of the thick cloud data and no-cloud data, then SIFT detection and feature point matching are applied in the overlapping region, furthermore, the exact matching point pairs can be extracted with proper strategy. Based on these exact matching point pairs and the quadratic polynomial model, the rectified image can be obtained. Meanwhile, thick cloud regions are detected by the algorithm of multispectral image analysis, and then the images of thick cloud regions are replaced with the corresponding regions of the rectified image. Finally, radiance differences are eliminated for image visual effect. Experiment results demonstrate that the proposed method can effectively remove thick cloud from MODIS image, which can satisfy the demand of post-processing for remote sensing imagery.