Automatic Building Extraction Based on Region Growing, Mutual Information Match and Snake Model

Abstract

Considering building extraction from high resolution remote sensing image was very difficult because of the complexity of land covers and the complication of building structures, in this paper, we proposed a new method for automatic building extraction based on improved region growing, mutual information match and improved snake model. Our work included the following four aspects. Firstly, we proposed a new method of noise reduction based on wavelet transformation and the Butterworth low-pass filter. Our scheme avoided the difficulty of threshold selection and could reduce the noises adaptively. Secondly, we proposed a new method of seed extraction based on scale, gradient and edge information. The true seeds which were relevant to targets could be extracted precisely. Thirdly, for homogeneity regions produced by region growing with extracted seeds, we defined three conditions to extract building templates with the shape of regular rectangle based on shape features. Fourthly, we proposed a method to extract candidate building regions based on mutual information match. Building contours were determined accurately based on improved snake model. According to the experiment result, our method can significantly improve the accuracy of building extraction, and almost all the buildings are extracted correctly.