Elaboration on a new RGB-split method to enhance building extraction in satellite images

Image processing became necessary in various scientific applications and different research fields, especially satellite imaging.In the field of remote sensing, which is our area of interest, several researchers have developed classification and segmentation methods that are very useful.However, those applications are limited regarding the complexity and diversity of satellite images.In this paper, we propose an original method for detecting buildings in RGB satellite images.The idea is to treat the three RGB matrices separately to accurately detect the pixel intensity variations, which provides better detection of building contours.Our method is mainly based on mathematical morphology operators.The method is a hybridization of two methods based on mathematical morphology which are the Hit or Miss Transform and the Top Hat, the Hit or Miss Transform detects all buildings because of its robust precision in detecting segments, after applying the HMT we apply the Top Hat to refine the segmentation result and finally detect clearly all building in the satellite image.We applied our method on several images from many datasets mainly Ikonos images, and Sentinel-2, the results of our method application gave great results with a Precision that exceeds 95%, Recall across 89%.