Journal of Geo-information Science, Volume. 22, Issue 2, 258(2020)
While extracting roads from high-resolution remote sensing imagery, shadow shielding is a main factor causing roads missing or defects, which could lead to difficulties for automatic road extraction. Therefore, developing methods for shaded road extraction with strong applicability has a great significance in map data production and research of geographical data. Traditional methods, such as the shadow coefficient amendment method, are difficult to remove the shadows of plants and buildings, and they would undermine the integrity of extracted roads. So, this paper proposed a feasible approach to extracting shaded roads based on brightness compensation and a high-performance segmentation method. First, after image preprocessing, a threshold segmentation method in HSI space was used to obtain the shadow area. Second, a combination of blue components suppression in the RGB space and divided linear strength was applied to enhance the pixel points in spatial domain and recover the information of the shaded areas, which made the difference between shaded roads and surrounding areas more obvious. Shaded roads were extracted by an efficient segmentation algorithm, and unshaded roads were calculated by K-means clustering segmentation. The initial value of clustering was based on color distribution in the HSI space. To ensure the integrity and details of extracted roads, the morphology method and contour repair algorithm were introduced into the extraction process after rough roads mergence. Results show that this method could extract shaded road successfully. For suburban roads, the integrity of extracted shaded roads was 96.84%. For urban roads, the accuracy was also higher than 80%. Compared with traditional methods based on the threshold segmentation in HSI, this method decreases the fragmentation of road patches while extraction, and keeps the integrity of the roads. This approach could be used for smart manufacturing and mapping of internet map data in high-resolution remote sensing imagery.
Get Citation
Copy Citation Text
Huixin HE, Junfu FAN, Wenhe CHEN, Yuke ZHOU, Peng ZHANG, Xiao YU.
Received: May. 29, 2019
Accepted: --
Published Online: Sep. 16, 2020
The Author Email: FAN Junfu (fanjf@sdut.edu.cn)