| The correlations between a material's microstructure and its physical behavior are not always fully understood. The development of tools for the three-dimensional characterization of material microstructures, the automation and the quantification of these methods is investigated. The watershed transform and stochastic segmentation algorithms are applied to four different structures obtained using different types of serial sectioning. A nickel based superalloy, IN100 and a beta titanium structure, Ti-8823, recorded using Focused Ion Beam (FIB) serial sectioning; a beta processed titanium structure recorded using the Robo-Met.3D system; and a biological data set, an Arabidopsis cotyledon, recorded by optical sectioning via confocal microscopy. Results on the segmentation of the grains in IN100 data set, the segmentation of laths in a Ti-8823 data set, the segmentation of the basket weave colonies of a titanium microstructure and the determination of different cell types in a plant cotyledon are presented. A method, based on mutual information, to quantitatively describe the quality of the segmentation of microstructures and to compare different segmentation algorithms is introduced. |
