Road extraction based on dynamic scale MSR and improved Canny in vague aerial images

Abstract

Road extraction in vague aerial or remote sensing images is hard but significant. In this paper, a road extraction method for vague aerial images is studied. The method consists of four main algorithms: (1) a modified MSR algorithm to enhance a vague aerial road image, in which, the logarithm transform is applied first, then the scales for Retinex are determined in light on the image depth by using the method of dark channel prior; (2) the enhanced image is roughly segmented based on the improved Canny edge detector with edge region expansion, then, edge image is thresholded by Otsu; (3) a number of post processing functions in a binary image is applied for gap linking and noise removal; and (4) the extracted roads are regulated by using the road shape features. In experiments, a number of vague aerial images of multiple roads were selected, and the algorithm comparison results show that the studied method is satisfactory for the road extraction in a vague aerial image, and it is much better than other traditional algorithms.