Two-Stage Algorithm for Segmentation of Satellite Images

Abstract

A two-stage algorithm for segmentation of satellite images is proposed, which makes it possible to detect extended textural areas and small-sized objects. The method is based on the representation of multi-digit digital images by a set of bit images and the use of the mathematical apparatus of two-dimensional Markov chains. Transition probabilities for a two-dimensional Markov chain and brightness characteristics are used as textural features of extended objects. The selection of small-sized objects is performed on the basis of an estimate of the amount of information using a mathematical model of a two-dimensional Markov chain. To reduce computational resources, the evaluation of features is carried out using binary images of the highest, most informative, image digits. The average accuracy of segmentation of extended areas according to the F-measure metric is 68.9%, for small-sized objects is 47.5%.