High-resolution remote sensing imagery provides an important data source for ship detection and classification.
However, due to shadow effect, noise and low-contrast between objects and background existing in this kind of data,
traditional segmentation approaches have much difficulty in separating ship targets from complex sea-surface
background. In this paper, we propose a novel coarse-to-fine segmentation strategy for identifying ships in 1-meter
resolution imagery. This approach starts from a coarse segmentation by selecting local intensity variance as detection
feature to segment ship objects from background. After roughly obtaining the regions containing ship candidates, a
shape-driven level-set segmentation is used to extract precise boundary of each object which is good for the following
stages such as detection and classification. Experimental results show that the proposed approach outperforms other
algorithms in terms of recognition accuracy.