Segmentation of multispectral images in optical metallography

The use of adequate multispectral image acquisition and segmentation procedures is essential for automatic phase identification in metallography. Spatial and supervised spectral classification methods are compared on a set of five different optical ore micrographs. Nearest neighbour classification complemented by a geodesic propagation filter appears to give the best results. This method is often able to rival with EDX mapping where cost and speed are essential parameters.