Number estimation of small-sized ships in remote sensing image based on cumulative projection curve

Ship detection is an important stage for the sea- area surveillance and many algorithms have been proposed for dealing with such tasks. Nevertheless, most of them are designed for large-sized ships and are not efficient for the small ones. In this paper, we present a novel method based on cumulative projection curve(CPC) to estimate the number of ships of small size. We firstly compute the Mahalanobis distance between each pixel of the image and the pixel intensities distribution of water, and then project these Mahalanobis distances to their near coastline vertically. The projected one-dimension curve is called cumulative projection curve. By doing this, each ship along the coastline will incur a fluctuating, the ship response, in the CPC. Thus, the number of ships can be estimated through the estimation for the number of ship responses in the CPC. This method simplifies the detection problem by converting a two-dimension problem to an one-dimension problem, and its efficiency is illustrated by the experimental results in the paper.