Object oriented approach for building extraction from high resolution satellite images

In this paper, an object oriented approach for automatic building extraction from high resolution satellite image is developed. Firstly, Single Feature Classification is applied on the high resolution satellite image. After that, the high resolution image is segmented by using the split and merge segmentation so that the pixels that are grouped as raster objects have probability attributes associated with them. Then different filters are applied on the image to remove the objects which are not of our interest. After filtering the segments, the output raster image is converted into vector image. After converting the raster image into vector image, the building objects are extracted on the basis of area. The cleanup methods are applied to smoothen the extracted buildings and also to increase the accuracy of extraction of buildings. Imagine Objective tool of ERDAS 2011 has been used. The approach is applied on three different satellite images. The extracted buildings are compared with the manually digitized buildings. For one satellite image it has picked up all the buildings with a slight change in the area of footprints of buildings. Only one patch of road is extracted as a building. For the other two satellite images, the overall accuracy is low as compared to the first satellite image. Some patches of road and ground are also extracted as buildings. The branching factor, miss factor, building detection percentage and quality percentage were also calculated for accuracy assessment. Nonetheless, the overall accuracy of building extraction with respect to area was found to be 85.38% in a set of 66 buildings, 73.81% in a set of 94 buildings and 70.64% in a set of 102 buildings.