Shape priors for MAP segmentation of alloy micrographs using graph cuts

Abstract

State-of-the-art alloy development processes utilize computerized materials simulations relying on segmentations of alloy micrographs which indicate the arrangement of material precipitates. Automated alloy segmentation algorithms must properly account for abundant prior information regarding the shape of the precipitates, as precipitate size and shape correlate to important materials properties. We propose a novel way of constructing and enforcing shape priors within a maximum a posteriori (MAP) segmentation framework. After computing a preliminary alloy segmentation using matching pursuits, our algorithm uses it to define a shape prior for a MAP segmentation found using a min-cut algorithm. This algorithm was found to significantly outperform both the MAP segmentation obtained without the shape prior, and the matching pursuit segmentations.