Radial Casting Algorithm for Extraction of Man-Made Features from High Resolution Digital Satellite Imagery

Abstract

The extraction of man-made features from high resolution digital satellite imagery is an important step to underpin management of geo-information in any country. Man-made features and buildings in particular are required for various applications such as urban planning, creation of geographic information systems databases and generation of urban models. Manual extraction processes are expensive, labor intensive, need well trained personnel and cannot cope with high demand of geo-information and changing environment. This paper, presents a Radial Casting Algorithm (RCA) used to extract buildings from high resolution digital satellite imagery. The algorithm measures only a single point on an approximate center of the building on an image and the fine measurement is automatically determined. The algorithm is a modification from original snakes model developed by Kass et al whereby the external constraints energy term is removed which negatively affects the convergence properties of the contour to provide the ability of the snake contour to cope with high variability of buildings on an image. The algorithm was tested on three areas of different environment. The quantitative measures were employed to evaluate the accuracy, efficiency and capability of the algorithm which shows that the time of extracting a single building was reduced by 32 percent, the extraction rate was 92 percent and the Area coverage of extracted polygons was 98 percent.