Single Aerial Photo Cloud Removal

Due to non-uniformity,cloud in an aerial photo is even harder to remove than haze.This paper proposes a divide-and-conquer based method of cloud removal using a single aerial photo.Based on color attenuation prior,we design a veiling index that indicates the local concentration of cloud on the aerial photo.By this index,an image veiled by cloud was segmented into various regions of various extents of veiling.Each region was veiled by nearly uniform cloud,and thus subjected to commonly used assumption such as boundary constraint on transmission in image patches.Atmospheric light vectors in each region were estimated by modified color-lines model and then composed into a spatially varying airlight map for the whole image.Initial estimated scene transmission was further optimized by a weighted L1-norm based contextual regularization,which was used to obtain scene reflection image thereafter.We demonstrate the effectiveness of our cloud removal method on a number of example aerial photos containing cloud and compare our results with that of existing relevant methods.