| In recent years, very high resolution (VHR) remote sensing is one of rapidlydeveloping technologies in the field of remote sensing. With the advent of VHR satelliteimages, a rich, vast amount of spatial data is possible to obtain. Road feature is one ofimportant characteristics of VHR satellite images. Road extraction from these VHRsatellite images is one of important and fundamental image processing technologies,which has a very wide range of applications in real life. Although there are various roadextraction methods in the literature, road extraction still appears to be in their infancydue to many negative factors, such as the natural scene complexity, image noise, andthe extraction algorithm limitations, such that road extraction has remained achallenging research subject.In this thesis, multiple road features are explored and integrated to extract roadnetworks from VHR satellite images. The objective of the proposed method is toimprove road extraction accuracy and provide a new road extraction solution.Specifically, the research work includes the following aspects:(1) A new shape feature is designed to quantitatively illustrate road characteristic.The proposed shape feature can effectively describe the geometric characteristics ofroads. It can be used to filter nonlinear features to improve road extraction accuracy.(2) A smooth road centerline extraction algorithm is proposed. The algorithmfirstly relies on tensor voting to decompose the road network into separate roadsegments, followed by the road centerline extraction using the regression algorithm. Incontrast to the traditional algorithms, test results demonstrate that the proposedalgorithm does not produce spurs and retain the centerline smoothness.(3) This thesis presents a multiple-features-based road extraction method. Thismethod combines the advantages of road shape features, image segmentation, andimage classification. Experiments show that the algorithm can produce accurate andstable road extraction performance. |
PDF Full Text Request