Recent Methods for Reconstructing Missing Data in Multispectral Satellite Imagery

One of the major limitations of passive sensors is their high sensitivity to weather conditions during the image acquisition process. The resulting images are frequently subject to the presence of clouds, which makes the image partly useless for assessing landscape properties. The common approach to cope with this problem attempts to remove the clouds by substituting them with cloud-free estimations. The cloud removal problem can be viewed as an image reconstruction/restoration issue, in which it is aimed at recovering an original scene from degraded or missing observations. Two cloud removal approaches are detailed and discussed in this chapter. The first one is a single-channel method for the reconstruction in a sequence of temporal optical images. Given a contaminated image of the sequence, each area of missing measurements is recovered by means of a contextual prediction process that reproduces the local spectro-temporal relationships. The second approach exploits the Compressive Sensing (CS) theory, which offers the capability to recover an unknown sparse signal with a linear combination of a small number of elementary samples. The two reconstruction approaches are evaluated experimentally on a real multitemporal multispectral remote sensing image.