Markov Random Field based microstructure reconstruction using the principal image moments

Abstract

The present work addresses the microstructure reconstruction of forged and additively manufactured materials using Markov Random Field (MRF) approach and the principal of image moments. The MRF based reconstruction is performed for the experimental samples to predict the spatial evolution of the microstructures on a larger scale. To achieve a high-fidelity statistically-equivalent representation for the original image, the synthesized samples are assessed according to their global and local level features. The global parameters measure the averaged properties of the microstructure image. In particular, they are defined as the distance to the image centroid and the eigenvalues of the covariance matrix of the normalized central moments. The local features are associated with the grain-level properties, including grain size and shape which are computed using the image moment values. With our presented approach, we not only demonstrate a computational framework for the spatial reconstruction of microstructures but also guarantee the statistical equivalency of the synthesized samples to the original microstructure images. • The microstructure evolution in large length-scales is predicted using the Markov Random Field approach. • The image moments are used to compare the synthesized microstructures with experimental samples in global and local levels. • The first Hu moment is observed to be dominant over other moments for capturing the grain shapes. • The uncertainty of the experimental images is found to be more effective in the local-level.