Thin Cloud Removal Using Local Minimization and Logarithm Image Transformation in HSI Color Space

In observation of land information using satellite images, clouds are one of the most serious obstacles due to their opacity property which can block the visibility of ground objects and can also be blended with the underlying details. Hence, retrieval the actual information covered by clouds is frequently necessary. In this paper, we propose a novel method to remove clouds by taking an advantage of HSI color space instead of directly removing clouds in RGB color space. The proposed method uses a concept of dark channel prior method to estimate the cloud appearance called the scattering light and perform a subtraction in only the intensity channel to avoid an effect to the original color and also enhance the intensity with gamma correction to recover some information accidentally removed from the previous step and restore obscure details distorted by clouds. Furthermore, since clouds involve in both intensity and saturation channel, we increase the saturation that was reduced as a result from clouds by using logarithm image transformation as well. From the results, the proposed method can remove clouds that are not extremely opaque and preserve the actual information such as color and texture due to the higher contrast gain in the experiments comparing to the results obtained from other single-image methods.