Island Road Centerline Extraction Based on a Multiscale United Feature

Abstract

Accurate island road centerlines are important to tourism planning, resource development, and other applications. However, high-resolution island imaging is highly detailed with complex features, which increases the difficulty of road centerline extraction. To address the problem, we propose a method based on a multiscale united feature, which can improve the extraction of road centerlines in high-resolution island images. First, a biological simulation saliency algorithm is used to determine the island extent. Then, the spectral, geometric, and texture features of roads in high-resolution images are fully utilized, and a multiscale united feature is generated by the effective combination of the features to enhance road characteristics. Road regions are extracted by the implementation of the united feature. Finally, with the advantage of reconstruction ability within strong noise, a tensor voting algorithm is applied to refine the road regions, and then, the road centerline is extracted. The following conclusions are obtained by comparing the proposed method with three state-of-the-art methods: For a study area in which roads present clear characteristics and are less covered, the proposed method exhibits better road centerline extraction performance, and its completeness, correct rate, and quality index are better than the methods to which it was compared; the same was true for a study area with complicated road distribution. The proposed method can be applied to road centerline extraction of high-resolution island images.