CLASSIFICATION BASED MARKER SELECTION FORWATERSHED TRANSFORM OF HYPERSPECTRAL IMAGES

A new method for segmentation and classiſcation of hyper-spectral images is proposed. The method is based on a pixel-wise classiſcation followed by selection of the most reliable classiſed pixels as markers for watershed segmentation. Fur-thermore, each marker deſned from classiſcation results is associated with a class label. By assigning the class label of each marker to all the pixels within the region grown from this marker, a spectral-spatial classiſcation map is obtained. Experimental results are presented on a 200-bandAVIRIS im-age of the Northwestern Indiana’s Indian Pine site. The devel-oped segmentation and classiſcation scheme signiſcantly de-creases oversegmentation, improves classiſcation accuracies and provides classiſcation maps with more homogeneous re-gions, when compared to pixel-wise classiſcation or previ-ously proposed spectral-spatial classiſcation techniques. Index Terms — hyperspectral images, segmentation, classiſcation, watershed, marker selection 1