Filtering to remove cloud cover in satellite imagery

The possibility of filtering light cloud cover in satellite imagery to expose objects beneath the clouds is discussed. A model of the cloud distortion process is developed and a transformation is introduced which makes the signal and noise additive so that optimum linear filtering techniques can be applied. This homomorphic filtering can be done in the two-dimensional image plane, or it can be extended to include the spectral dimension on multispectral data. The three-dimensional filter is especially promising because clouds tend to follow a common spectral response. The noise statistics can be estimated directly from the noisy data. Results from a computer simulation and from Landsat data are shown.