Visual Saliency Based Ship Extraction Using Improved Bing

Abstract

In this paper, a novel ship extraction algorithm is proposed to acquire the precise segmentation. First, Binarized Normed Gradients (BING) is improved to locate the potential ship regions according to the remote sensing application. Second, the visual saliency computation by combining Hypercomplex Frequency Domain Transform (HFT) and Phase Quaternion Fourier Transform (PQFT) presented to analyze the located regions. Third, segmentation on the computed saliency map is conducted to extract precise blobs of the ship candidates. Finally, the prior knowledge about ship is utilized to remove the false alarms in the discrimination stage. The experimental results of typical remote sensing images show that the algorithm is very effective.