A multilevel extraction approach of linear objects in high resolution SAR image

Extraction or recognition of linear objects in high resolution SAR images has been a focus of attention in SAR data applications. Although there are many approaches to extract straight lines, these approaches have more or less limits in their applications. In this paper, for the coarse-to-fine information integration at different scales, a multilevel extraction approach aimed at improving linear object extraction precision, robustness and computational efficiency is presented. Firstly, the effect of the image scale to linear feature extraction in SAR image is analyzed. Then, a multilevel algorithm which can extract linear feature fast and locate it accurately is presented. At last, the approach is tested by the runway extraction, and the results are analyzed.