ISSN 0474-8662. Information Extraction and Processing. 2017. Issue 45 (121)
Home Back to issue

Method of features construction for remote sensing images based on the characteristics of random point fields

Kosarevych R. Ya.
Karpenko Physico-Mechanical Institute of the NAS of Ukraine, Lviv
Lutsyk O. A.
Karpenko Physico-Mechanical Institute of the NAS of Ukraine, Lviv
Rusyn B. P.
Karpenko Physico-Mechanical Institute of the NAS of Ukraine, Lviv
Korniy V. V.
Karpenko Physico-Mechanical Institute of the NAS of Ukraine, Lviv

https://doi.org/10.15407/vidbir2017.45.090

Keywords: random point fields, remote sensing, image segmentation, texture descriptors

Cite as: Kosarevych R. Ya., Lutsyk O. A., Rusyn B. P., Korniy V. V. Method of features construction for remote sensing images based on the characteristics of random point fields. Information Extraction and Processing. 2017, 45(121), 90-95. DOI:https://doi.org/10.15407/vidbir2017.45.090.


Abstract

Texture features are widely used in remote sensing image classification. In most cases they are extracted from grayscale images without taking color information into consideration. The texture descriptors, which consist of characteristics of random point fields formed for pixels of distinct intensity of grayscale and color band images are presented. The input image is divided into fragments for the elements of each of which the histogram is constructed and their local maxima are determined. Size of fragments are chosen depending on image resolution. For each of the intensity of the dynamic range of the image, a random point field, as a set of geometric centers of fragments, is formed. By the formed configuration, each field is classified as cluster, regular or random. To form a description of image elements a distribution of the number of field elements for each intensity and fragment is constructed. Separately, the vectors of the point field element for each intensity in the image fragment and the point field element for the selected intensity are formed. Experimental results demonstrate that proposed descriptors yield performance compared to other state-of-the-art texture features.


References