Comparative analysis of various thresholding techniques on TerraSAR-X images in the presence of speckle noise

This paper compares several methods of thresholding applied to TerraSAR-X radar images in the presence of speckle noise and a method based on the use of the local extremas of the histogram. The methods used are Otsu´s, Valley-emphasis, maximum entropy, fuzzy sets, Yagar´s measure of fuzziness and histogram clustering. The local minimum of the histogram provides a good threshold candidate for global thresholding in ship detection in the case of moderate signal noise but fails for high signal noise and variable illumination. The methods are tested on sample images to evaluate their performance in object detection in real TerraSAR-X datasets.