Automated image segmentation for analysis of materials microstructures
Abstract
Automated image segmentation has become a core part of materials analysis, for characterizing the three-dimensional (3D) microstructure of materials. The rate of data collection has been growing rapidly in recent years due to improvements in automatic image acquisition, so image analysis involving much human intervention becomes infeasible. We address the above issues by providing a new image modeling framework, the Markov Chain Markov Random Field (MCMRF) model for automatic image segmentation of image sequences. The MCMRF model may incorporate the spatial relationship of the z direction to build a more appropriate prior model for image segmentation of image sequences of 3D microstructure. Experimental results show improved accuracy of segmentation over its 2D counterpart. The second part of this thesis presents a flexible metric to evaluate image segmentation using the Weighted Rand Index (WRI). We incorporate the edge map and curvature into the evaluation framework, although other features could also be integrated using appropriate feature extraction. With the 3-way contingency table, a weighted measure of similarity can be computed to compare the segmentation results from a segmentation algorithm against the ground truth. The evaluation of real microscope images of materials will be presented.
Degree
Ph.D.
Advisors
Comer, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.