Road Extraction from VHR Remote-Sensing Imagery via Object Segmentation Constrained by Gabor Features

Li Chen,Xiao Xie,Qing Zhu,Han Hu,Haowei Zeng

doi:10.3390/ijgi7090362

Abstract

Automatic road extraction from remote-sensing imagery plays an important role in many applications. However, accurate and efficient extraction from very high-resolution (VHR) images remains difficult because of, for example, increased data size and superfluous details, the spatial and spectral diversity of road targets, disturbances (e.g., vehicles, shadows of trees, and buildings), the necessity of finding weak road edges while avoiding noise, and the fast-acquisition requirement of road information for crisis response. To solve these difficulties, a two-stage method combining edge information and region characteristics is presented. In the first stage, convolutions are executed by applying Gabor wavelets in the best scale to detect Gabor features with location and orientation information. The features are then merged into one response map for connection analysis. In the second stage, highly complete, connected Gabor features are used as edge constraints to facilitate stable object segmentation and limit region growing. Finally, segmented objects are evaluated by some fundamental shape features to eliminate nonroad objects. The results indicate the validity and superiority of the proposed method to efficiently extract accurate road targets from VHR remote-sensing images.

Highlights

Since roads are a principal part of modern transportation, managing and updating road information in the Geographic Information System database is of great concern [1]
Some factors that contribute to the difficulty of high-resolution road extraction are: increased data size and superfluous details with progressively higher resolutions, which means more noise interference and processing time; shelters, such as vehicles and trees on the roadside or shadows of artifacts and buildings, vehicles’ location can be identified by eavesdrops, it breaks the users’ trajectory privacy [4]; the phenomenon of similar objects possessing varying spectra while different objects possessing the same [5], which can cause wrong segmentation results by a region-based method; the necessity of finding weak road edges when the spectral representation of the road surface is similar to the background; and, sometimes, the demand for fast acquisition of road information when facing a crisis [6]
Buslaev and Seferbekov use the fully convolutional neural network of U-Net to extract road networks [10], while a single patch-based convolutional neural networks (CNN) architecture is proposed in Reference [11]

Summary

Introduction

Since roads are a principal part of modern transportation, managing and updating road information in the Geographic Information System database is of great concern [1]. Knowledge-based methods detect roads by using higher information, e.g., a vison-based system, proposed by Poullis and You [8], uses Gabor filtering and tensor voting for geospatial-feature classification and graph cuts for segmentation and road feature extraction. Buslaev and Seferbekov use the fully convolutional neural network of U-Net to extract road networks [10], while a single patch-based CNN architecture is proposed in Reference [11]. Despite these methods showing superior results, their inadequacy in keeping weak, tenuous edges could diminish the completeness of edge information

Methods

Results

Discussion

Conclusion