Sea detection on high-resolution panchromatic satellite images using texture and intensity

In this paper, a two-stage sea-land mask detection algorithm on high resolution panchromatic images is proposed. An initial mask is generated using texture features in the first stage and this mask is refined by using intensity values in the second stage. Image is divided into windows and the Local Binary Patterns (LBP) histograms, evaluated at each window, are modelled using the sea and land sample spaces obtained by the altitude information which has very low resolution compared to the image. These models are utilized for graph cut segmentation algorithm to generate the initial mask. Output mask is generated by thresholding the geodesic distance to the eroded initial mask, calculated on the enhanced and filtered image. Test results obtained on satellite images showed that the proposed algorithm is capable of detection of sea with a high accuracy rate.