Road Centreline Extraction of High-Resolution Remote Sensing Image with Improved Beamlet Transform and K-Means Clustering

The extraction of road centreline from a remote sensing image is an important part of obtaining and updating basic geographic information data. In view of the shortcomings of traditional road centreline extraction algorithm, such as the presence of burrs, lengthy time consumption and ineffective extraction of the centreline around intersections, we propose an algorithm based on an improved beamlet transform combined with K-means clustering to extract the road centreline of high-resolution remote sensing images. The proposed method includes the following steps. Firstly, K-means clustering is used to identify the road area. Secondly, the improved beamlet transform algorithm is utilised to extract the road centreline. Reducing the beamlet base and simplifying the beamlet dictionary reduces the algorithm complexity. By adding energy threshold and direction constraint, each binary block is guaranteed to contain only one optimal base that avoids the overlapping and missing of the beamlet base. Lastly, a connection algorithm is designed to connect the breaks of some road centrelines. Experimental results show that the proposed algorithm can accurately extract the road centreline, effectively form high-resolution remote sensing images and had good anti-noise ability. Compared with traditional beamlet transform, the proposed beamlet transform achieves better results.