Coastline Detection from Remotely Sensed Imagery: Development of a Methodology Based on Advanced Smoothing Techniques and the Canny Edge Detector

摘要

Coastline extraction is a task of fundamental importance for coastal zone management and its accurate detection has greatly concerned the scientific community for more than a decade. In this paper an algorithm for the semi-automatic coastline detection from remote sensing data was designed and developed. The algorithm consists of three processing steps. In the first step, certain edge-preserving filters were applied for data enhancement, smoothing and simplification. Then, image edges were detected based on the Canny edge detector and resulted pseudo-edges or undesirable ones were eliminated based on certain processing techniques, namely measurements of entropy, K-means clustering and thresholding. In the final step, a post-processing scheme was implemented in order to deal with the connectivity of the detected coastline based on mathematical morphology operators. The developed methodology was applied on six high resolution QuickBird and WorldView-2 images, one medium resolution ASTER image and two high resolution TerraSAR-X radar data. A quantitative and qualitative evaluation was performed using the standard measures of completeness, correctness and quality. The comparison with the ground truth data from photo-interpretation demonstrated the effectiveness of the developed algorithm.