Vol. 13, issue 02, article # 3

Abstract:

We propose an approach to deconvolution of satellite images of the Earth's surface recorded under conditions of atmospheric distortions. The specific feature of this approach is that the point spread function (PSF) used in the linear model of reconstruction is unknown and thus it ought to be estimated first. To do this, the image itself is used together with the information that the observed scene contains objects having certain brightness contrast. We use the Gumbel distribution of extrema as the stochastic model of degraded image fragments with high gradient. Otherwise, the Johnson curves are used in the description. The Bayes decision rule that uses these distributions isolates extremal gradients. The variations of brightness along the directions of gradients in the blurred part of the image serve a basis for reconstructing the PSF. The image itself is recovered using a standard approach. Illustrations to the PSF identification and image reconstruction are also given in the paper.