Application and further development of advancedimage processing algorithms for automated analysis ofserial section image data

摘要

Several automated algorithms are presented for the segmentation of featuresof interest from microstructure images acquired with modern high-throughputelectron microscopes. Specifically, the maximization of posterior marginals(MPM) segmentation technique, originally developed for computer visionapplications, is applied towards automated segmentation of microstructuralimages from Ti- and Ni-alloy systems. The MPM technique classifiesimage pixels according to the most probable class to which they canbelong. Three derivatives of the MPM algorithm are introduced and assessed:expectation maximization MPM (EM/MPM), EM/MPM with simulatedannealing (EM/MPM/SA) and vector EM/MPM/SA. Example applications ofall three approaches are given. The EM/MPM model allows for automatedsegmentation of a laths in a Ti-6242 sample and primary y' in an IN100superalloy, but has difficulty accurately locating the boundaries betweenregions. The EM/MPM/SA algorithm involves a gradual increase in theinterface capillarity during segmentation and allows for pixel accuracydetermination of boundaries between phases. The vector EM/MPM/SA methodis capable of simultaneously segmenting a series of images acquired withdiffering imaging conditions. The limitations of the algorithms are discussedas well as potential future modifications.