Reconstruction of cloud-contaminated multitemporal optical images with a contextual prediction method

The aim of this work, is to propose a new single-channel method for the reconstruction of areas obscured by clouds in a sequence of temporal optical images. Given a cloud-contaminated image of the sequence, each area of missina measurements is reconstructed by means of an ensemble of contextual linear predictors that reproduce the local spectro-temporal relationships between the considered image and an opportunely selected subset of the remaining temporal images. Each predictor of the ensemble is trained in an unsupervised way, over a local multitemporal region that is spectrally homogeneous in each temporal image of the selected partial sequence. In order to obtain such regions, each temporal image is locally classified by, means of an unsupervised classifier based on the Expectation-Maximization (EM) algorithm. To illustrate the performance of the proposed method, an experimental analysis on a sequence of three temporal images acquired by the Landsat-7 ETM+ sensor is reported.