Performance testing of selected automated coastline detection techniques applied on multispectral satellite imageries

Coastline detection has been of major interest for environmentalists and many methods have been introduced to detect coastline automatically. Remote Sensing techniques are the most promising ones to deliver a satisfactory result in this regard. In our study, the objective was to retrieve performance level of certain image processing techniques vigorously used for the purpose to delineate coastline automatically and they were tested against two images acquired almost on the same period by LISS III and LANDSAT ETM+ sensors. The algorithms used in the study are Water Index, NDVI, Complex Band Ratio, ISODATA, Thresholding, ISH Transfirmation techniques. Accuracy of the shoreline detection by classifying the image in land and water has been tried to be estimated in three ways, firstly with comparison to the visually interpreted high resolution google earth image, secondly field collected GCP data of reference points of classes and thirdly the raw image itself. But problem in temporal disparity caused the constraint doing accuracy assessment from the first two reference data and maps along the coast. As a whole although four techniques among six, show satisfactory results namely density slicing, ISODATA classification, Water Index and ISH transformation technique, in the case of LISS-III and ETM+, Water Index (with kappa value being 0.95 for LISS-III and 0.97 for ETM+) and Intensity-Hue-Saturation transformation techniques give better performance. Sensor to sensor variation might have introduced certain differences in shoreline detection in images of same season with similar tidal influence.