|Articles and journals | Tariffs | Payments | Your profile|
Segmentation of satellite images based on super pixels and sections on graphs
Abstract.The study is devoted tp algorithms of segmentation of satellite images for various systems of technical vision. For the segmentation of images authors use sections on graphs. Preliminary segmentation is performed based on the minimal spanning tree to improve performance. When describing the properties of super pixels, information about the height and color of the regions is taken into account. The height of the areas is calculated based on the stereo images. The color of segments is calculated on the basis of color invariants. All super pixels in accordance with their characteristics belong to the areas of buildings, grass cover, trees and shrubs, shaded areas, etc. The image is an undirected weighted graph, the nodes of which are segments of the image. The weights of the vertices of a graph are numbers that determine the membership of a certain class. To divide regions into clusters, the method of cuts on graphs is used. The novelty of the study is the algorithm for segmenting satellite imagery based on super pixels and graphs. The segmentation time on the basis of the developed algorithm decreases several times in comparison with the method of cuts on graphs. The developed algorithm is used to allocate buildings to images. Comparison of the developed algorithm with existing approaches of building allocation is shown, its advantages are shown. Examples of the operation of the algorithm are given by the authors of the article and the results of the research are described.
Keywords: pattern recognition, computer vision, image processing, satellite imagery, graph cuts, spectral graph theory, superpixels, image segmentation, scene analysis, color invariants
Article was received:16-03-2018
This article written in Russian. You can find full text of article in Russian here .