Hybrid Algorithm for Filling in Missing Data in Electron Backscatter Diffraction Imaging

Emmanuel A. Atindama, Huston Wilhite, Cody Mattice, Gunay Dogan, Prashant Athavale

Abstract:

In this study, we present a novel hybrid algorithm for filling in missing orientation values in Electron Backscatter Diffraction (EBSD) maps, which are instrumental in characterizing microstructures of polycrystalline materials.
Traditional exemplar-based and machine-learning methods, which were originally designed for natural images captured with conventional cameras in macroscopic environments, typically perform poorly on EBSD images due to the distinct nature of EBSD data and grain geometries.
To address this gap, we adapted a classical exemplar-based inpainting algorithm and a partial convolutional neural network-based method to the unique requirements of EBSD data.
Each adapted method, however, presents its own set of strengths and limitations.
To overcome these limitations, we propose a novel hybrid inpainting algorithm that integrates modified machine learning and exemplar-based methods.
Our approach begins by using a tailored deep learning model to produce an initial approximation of missing regions, followed by refinement using an adapted exemplar-based algorithm, ensuring the preservation of critical grain boundaries and structural integrity.
We validate our method on 10,000 synthetic EBSD images generated with DREAM.3D software and demonstrate that the hybrid algorithm outperforms standalone techniques in both accuracy and visual coherence.
This advancement results in high-quality restorations that are more reliable for material analysis.
Furthermore, the hybrid approach is adaptable to broader inpainting problems, extending its applicability beyond EBSD maps to other orientation-based datasets.