Object-Oriented Port Detection Based on Mean Shift Segmentation

In this paper, we propose a new method for port detection using remote sensing image. This technique is a combination of image segmentation and object-oriented detection. We use mean shift method to initially segment the remote sensing image. Based on the initial segmentation map we build the spatial adjacency relation of the regions. Significant department has been observed in object-oriented detection. Generally, ports have inherent characteristic of half close region of seawater. Convex polygon created by centers of land region along the coastline. The land region is controlled within a certain range. Port detection is implemented according to water areas within the convex polygon. The advantages of this method are the robustness and its affine invariance with respect to translation, scaling, rotation and skewing. The effectiveness of this algorithm is demonstrated using two remote sensing images.