Sea-Land Segmentation Based on Template Matching

Abstract

In this paper, a novel sea-land segmentation method based on template matching is proposed. Firstly, we use OTSU to get coastline curve from harbor image. Subsequently, the curvature integral of the coastline curve is calculated as our matching signature, and matched parts between the two curves are found by using normalized cross-correlation. Finally, affine transformation is estimated and applied, thus completing the sea-land segmentation in harbor image. In order to research the limit capacity of template matching over the image resolution and relative angle between template image and harbor image, experiments on Google Map dataset had been made, and the boundary condition on resolution and angle was analyzed based on the experimental results.