Improvement of Buildings Detection Based on Adaptive Thresholding in Satellite Images

In this paper, a combined method to detect buildings from satellite imagery is presented. This method is based on combining data obtained by the local feature vectors and decision-making by applying adaptive thresholding the estimated probability distribution function, is conducted. Local Features serve as observations and location of buildings are used as joint random variables in order to estimate the probability density function. Then the locations of buildings are determined by considering modes of estimated probability density function and extracted features. To evaluate the efficiency of proposed method some satellite imagery of northern Tehran is used. Satellites North of Tehran building images are used in order to evaluate our proposed method. Tested images have different spatial contrast and resolution. Furthermore, tested buildings contain variety of characteristics which allows us representing our simulation with sufficient diversity. Experimental results of 32 different images in Tehran have shown that proposed method can be detected existing buildings in satellite imagery with fewer errors and more accurately.