Coastline detection in SAR images using wavelet packets, Multiscale Segmentation and a Markov Random field regularization

There has been growing research in the area of coastline detection using SAR images over the past few years. In this paper we propose a novel coastline extraction method based on wavelet packets, multiscale segmentation and a Markov Random Field regularization. Numerous spatial domain classical algorithms currently failed in the discrimination of Water and Ground when the contrast within the pixels values is low. Suitable wavelet packets informations features provides a good tool for distinguishing between textures. Utilizing the inherent tree structured of wavelet packets, a multiscale texture segmentation based on the fuzzy C-means algorithm is performed at different scales. The aboved multiscale segmentation are fused using a Markov Random Field regularization in the features domain for the final extraction of the coastline. The experimental results show the performance of the method, we can visualy evaluated the improved quality of the coastline extracted compared to classical algorithm based on image domain. Somes results are presented with ERS SAR images.