SEGMENTATION OF NATURAL AND ANTHROPOGENIC OBJECTS FROM PANCHROMATIC SATELLITE IMAGES USING STATISTICAL TEXTURE FEATURES
E.V. Dmitriev1,2, T.V. Kondranin2, S.A. Zotov2
1Marchuk Institute of Numerical Mathematics of the Russian Academy of Sciences, Moscow, Russia
2Moscow Institute of Physics and Technology, Moscow, Russia
Keywords: remote sensing, pattern recognition, texture features, thematic processing, classification, segmentation
Abstract
The problem of segmentation of natural and anthropogenic objects from panchromatic satellite images of very high spatial resolution (< 1 m) using texture analysis is considered. The effectiveness of various statistical methods for extracting texture features is analyzed. Based on the results of numerical experiments represented in this paper, we have selected methods that make it possible to segment the main types of natural and anthropogenic objects, as well as various structures of the forest canopy, with high accuracy (> 95%). We proposed the TTSPCA method, which allows the joined use of the most informative features extracted by different statistical methods. The results of numerical experiments show that the method proposed has higher texture segmentation accuracy (> 99%) in comparison with the standard texture extraction methods considered in this paper.
|
