A cloud removal approach for aerial image visualization

Partial cloud cover is a serious problem in synthesis of virtual scene and terrain models. The problem can be mostly resolved by mosaicking the cloud areas with the cloud-free areas in other temporal images. In this study, a complete approach, including image enhancement, cloud detection, cloud areas mosaicking, and image feathering, is proposed to generate cloud-free images from multi-temporal satellite images. At first, all original images were enhanced by the intensity histogram equalization in the proposed exact HSI (eHSI) color space, and the gamut problem could be avoided. Second, a simple intensity thresholding method was used to extract all cloud pixels, and then a difference checking was used to release the fixed-flat-white land covers in the eHSI color space. Third, all images were divided into grid zones, and then the cloud zones in the base image were replaced with same-location, cloud-free zones from other temporal images. Finally, a color uniformity method and a pyramid multiscale feathering method were used to feather the replaced cloud zones to generate realistic, cloud-free satellite images. Based on the proposed complete approach, not only were the clouds removed in the resulting images, but also, the resulting images show proper brightness and saturation. Moreover, the effect of the proposed methods is demonstrated and compared with other existing methods.