A Shape Constraints Based Method to Recognize Ship Objects from High Spatial Resolution Remote Sensed Imagery

摘要

Automatically extracting moored ships from high spatial resolution remote sensed imagery is more difficult than offshore ships because their spectrum values and textures are very close to the harbor. For this reason, different routes are designed and applied to extract the two kinds of ships. Our whole method can be divided into three main steps: 1) extract water polygons with histogram segmentation, 2) extract holes in the water polygons with morphological operations as the possible offshore ships, and extract possible moored ship with the identification of the salients to sea along the water boundary, and then 3) screen real ships out of these possible ships with more shape constraints. A case study is carried out on Spot 5 imagery to validate our method.