SAR Images Coastline Detection Algorithms Based On Active Contour Models

Coastline detection from the Synthetic Aperture Radar (SAR) Image plays an important role in the fields such as marine management, automatic navigation, ship target recognition and coastal map updating etc.Because of large variety in ocean area reflection information, complex landform and the strong speckle interference factors, it is difficult to detect the coastline from SAR image. According to the different kinds of SAR images, three modified active contour model algorithms have been provided for coastline detection. The main work of this paper as follows:(1) A modified Chan-Vese model based on regularization term is proposed for coastline detection.Utilizing CV model to detect coastline from radarsat images, many isolated small areas occurred due to strong speckle noise. To overcome such problem, the basic principle of CV model and its regularization term are analyzed firstly, then a modified regularization term based on a ROEWA operator is presented. Subsequently, a detailed theoretical derivation of the level set evolution equation and numerical calculation algorithm are provided. Finally, experiment results shows that the modified model can eliminate such isolated small areas caused by strong speckle noise. Meanwhile, the modified model has better performance than CV model in running time under low variance of speckle noise in Radarsat images.(2) A novel region-based active contour model dependent on inverse gaussian distribution is presented to detect coastline. In envisat images, when the ocean areas intensity tends to be homogeneity, while the land areas intensity tends to be inhomogeneity, the whole image intensity inclines to the low gray value. Assuming that the speckle distribution obeys the inverse Gaussian distribution with unit mean and radar backscatter is constant, a region-based energy functional is provided by a maximum likelihood principle. Then the level set function, the regularization term which smooths the contour, and the penalty term are introduced into this energy functional. Finally, a detailed theoretical derivation of the level set evolution equation and numerical solution algorithm are provided. Experiments results show that the proposed model has better performance contrast with existing Gamma model in Envisat images which meet the conditions.(3) A modified local region-based active contour model based on lognormal distribution is presented to detect coastline. In Radarsat and Envisat images, inhomogeneity in ocean surface often occurs due to the influence of wave, where some pixels in ocean areas demonstrate to be similar to those of lands. A modified local region-based active contour model is proposed, which combines the kernel function existing region-scalable fitting model and the hypothesis of lognormal distribution for SAR images. The principle of maximum a posteriori probability is utilized, and then level set function, the regularization term, and the level set penalty term are introduced in the energy functional. A detailed theoretical derivation of the level set evolution equation and numerical solution algorithm are provided. Finally, the effectienty of the modified model is verified by experimental comparison with Gamma model using the radarsat and envisat images which meet the conditions.