A STATISTICAL MODEL OF SEA CLUTTER IN PANCHROMATIC HIGH RESOLUTION IMAGES

From the perspective of developing a ship detection algorithm on optical imagery, a statistical model is developed to approximate histograms from high-resolution images of the sea surface. This model is developed using an empirical approach based on analysis of hundreds of images acquired on all the oceans of the planet. Several statistical distributions are selected in agreement with the state of the art in remote sensing of sea surface and ship detection. Thumbnails of different sizes are extracted from satellite images, their histograms are then calculated. The generated histograms are approximated by the probability density functions of the different statistical distributions selected. The least-squares method is used. Reliability of the models is tested by applying the KolmogorovSmirnov test and analyzing the sum of squared residuals in least-squares sense. Alpha-stable distribution is retained as the best among tested models. Texture and frequency descriptors are calculated and compared with Alpha-stable parameters to assess relations binders. Reliability of the models according to the sensors, the sea state is discussed.