A Method for Extracting Human Settlements from Remote Sensing Image Using Right Angle Corners Features

A method for extracting human settlements from high resolution remotely sensed panchromatic images using statistical characteristics of right angle corners is proposed in this paper. First, both right angle corners and straight lines are extracts from the images. Second, each right angle corner is optimized, based on the constraint of the lengths of two supporting straight lines. Third, a feature image is derived by projecting each right angle corner into its buffer zone. At last, human settlements are extracted by thresholding the feature image. Comparing the statistical characteristics based on the corners, straight lines and right angle corners without the constraints of the lengths of their supporting straight lines, the statistical characteristic based on right angle corners with the constraints of the lengths of their supporting straight lines is better to separate the human settlements from the others. This can decrease the negative effects of the roads, vehicles, the farmland ridges with both corners, and straight lines features. In addition, as compared with the existing methods for human settlement extraction using the PanTex Index, the proposed method is independent of the complexity of the textures, which can decrease the negative effects of complex textures.