Semiautomatic extraction of building information based on mathematical morphology

Abstract

An increasing number of applications of automatic extraction of building information have been taken in, such as in city plan, city development, military affairs, national defense, and high resolution remote sensing imagery provides these applications with new data sources. Based on differential morphological profile (DMP) algorithm in mathematical morphology, we introduce and develop a set of accurate and automatic scheme for figure information extraction, combining with MHN filter algorithm, region marking algorithm, area threshold segmentation algorithm and so on. First, the Maximum Homogeneity Neighbour Filter Method (MHN) was used to improve the quality of the image. Then, the derivative of the opening profile from the DMP algorithm of variable step size was in use for extraction of buildings' figure. At last, area marking and area threshold segmentation algorithm were introduced for post processing. DMP algorithm in classic definition requires constant step operator with different radius for differential. This means extraction of regular figure of buildings will get good results, but buildings with more complex shapes of will lapse by DMP. Aimed at the characters in the test field, we adopted a variable step size. By changing the constant step size for differential, our method can effectively detect not only buildings with regular shape, but also buildings with complex shape. In the end, accuracy evaluation was carried out for the extraction results in the test field and the precision reached 79.6%.