This dissertation work is concerned with the important scientific and applied problem of increasing the efficiency of methods for the preprocessing, analysis, and interpretation of multidimensional photogrammetric high spatial resolution images, including Earth remote sensing data.
A new object-oriented technology for multidimensional satellite imagery interpretation, which allows one to obtain classified images of the Earth surface, was developed. The developed technology allows one to analyze and identify individual objects of the Earth surface by calculating their features. To identify various types of objects, geometric, spatial, spectral, texture, and static features were proposed. At the segmentation stage, various segmentation methods that are used in multidimensional image decoding and analysis were applied to the input data: mean-shift segmentation, multiresolution segmentation, and K-means based segmentation. The analysis results show that multiresolution segmentation is the most efficient segmentation method for high spatial resolution multidimensional satellite images.
In the dissertation work, a shadow detection and removal technology for multidimensional high-resolution satellite imagery was developed. The problem of shadow removal consists in obtaining an image without shadows, i. e., the illuminance of the shadow regions must be the same as that of the rest of the image. For this purpose, the shadow regions must be processed. To implement shadow removal and to obtain a shadow-free image, the following problems were solved: shadow identification, shadow removal, and shadow segment border processing. In the developed technology, the shadow removal method is based on a shadow formation model that takes into account the physical principles of light scattering and reflection. The developed method of shadow removal was enhanced by adding a paired region search procedure and a shadow region border processing procedure. A search for paired shadow and non-shadowed regions that belong to the same type of surfaces improves the quality of the shadow detection and removal.
A method of geometric transformation of high spatial resolution aerospace imagery was refined. Today’s Earth remote sensing systems are equipped with scanner systems that can capture multidimensional images of the Earth surface. Because of the design features of scanner systems, the geometry of this type of imaging differs from the geometry of conventional photography. To make a correction by geometric transformation, ground control points and linear image objects were used. An automatic identification of image linear objects based on the developed decoding algorithm was proposed.
