Extraction of buildings using a fuzzy-morphological method

Buildings extraction is extremely important for many applications such as natural disasters and crisis management, urban planning and development, or telecommunication, etc. However, extracting automatically buildings from multispectral urban satellite images isn't an easy task because, in one hand, it often needs other technological expensive resources such as Lidar data. In the other hand, the urban environment is becoming more complex and heterogeneous in colours and shapes of its components. From that, we propose in this paper a fuzzy-morphological method applied on urban satellite images delivered alone. It is an hybrid method based on a combination of spectral and spatial pixel information to overcome the limitation of traditional methods. The spectral information is treated using fuzzy logic concept and a spatial one using morphological operations. It can be considered as a thresholding method which achieves the aimed goal according two steps: at first, it does a soft clustering by attributing each pixel to a cluster according to a membership degree, after that, it keeps only interesting regions “buildings”. Promising buildings extraction results are presented and to show even more the effectiveness of proposed approach, it has been compared to two other thresholding methods.