A Bilateral Texture Filtering Based Cloud Detection Method for VHR Satellite Images

New commercial satellites in orbit such as GeoEye-1, WorldView-3, and PlanetLabs can provide and update very high resolution (VHR) images of the Earth surface frequently. Such VHR satellite images contain a lot of fine spatial details and texture information, which pose challenges to existing cloud detection techniques. To avoid high demand for computing resource and over dependence on spectra, a bilateral texture filtering based cloud detection method is proposed in this paper. Firstly, the proposed method builds a significance map to divide the input image into noncloud regions and candidate cloud regions. Secondly, an optimal thresholding is calculated and used on the significance map to get a coarse result of detection. Then, the multiscale BTF is employed to capture the accurate detail map of the input image to remove the noncloud regions in the coarse result of detection. The final binary result is obtained by erode, dilate and guided feathering processes. The experiment is carried out on two sets of VHR satellite images. Subjective analysis and objective evaluations show that the proposed method works well for RGB color and grayscale images. It can produce high accuracy cloud detection results and outperforms some existing traditional methods.