MULTI-SPECTRAL IMAGE SEGMENTATION ALGORITHM COMBINING SPATIAL AND SPECTRAL INFORMATION

To segment multi-spectral images new methods are required which operate both in spectral and spatial domains and work with a high-dimensional data. We are presenting a new segmentation method that is built from standard statistical pattern recognition algorithms. It integrates spectral and spatial domain information by a combined classifier approach. We have studied the algorithm performance on real multi-spectral images of detergent powders acquired by the method of scanning electron microscopy. The use of apriori information for the segmentation of images with the similar spectral properties is investigated. The algorithm stability and the measure of segmentation quality are discussed as well. The proposed method appears to be a robust solution for the multi-spectral image segmentation. keywords: multi-spectral images, image segmentation, classifier combination, EM clustering