Image adaptive segmentation algorithm for unmanned surface vehicle targets

Abstract:

Considering the large contrast changing of surface targets and sea-sky background and the obvious difference of field depth, an improved image segmentation algorithm based on self-adaptive Mean-Shift is proposed. Spatial bandwidths are adaptively computed according to the estimation of gray distribution around the reference point; then the gray-level bandwidths are adaptively computed with a novel Bayesian theory in the corresponding windows; and finally adaptive segmentation is obtained. In the experiment, both the close and distant target frames, as well as target frames of different contrast, are extracted respectively from the surface vehicle video sequence. Compared with the traditional segmentation algorithm, experimental results prove that the proposed algorithm can effectively complete segmentation of surface target images.