An Automated and Unbiased Grain Segmentation Method based on Directional Reflectance Microscopy, Wittwer et al.

This repository contains the data and code necessary to reproduce the results presented in our publication. Abstract: Identifying individual grains from sectioned polycrystalline metals is a foundational task of microstructure analysis. However, traditional grain segmentation methods applied to optical micrographs may suffer from the lack of optical contrast between grains and require the manual selection of adjustable parameters to achieve acceptable segmentation results. We propose an alternative method which takes advantage of a multi-angle optical microscopy technique termed directional reflectance microscopy. By combining dimensionality reduction, similar-dissimilar classification, and multi-region merging of surface directional reflectance, our method enables fully automated and reliable grain segmentation of polycrystalline surfaces. We apply our method to metal samples with different crystal structures and grain orientation distributions. Our results suggest applicability of the method to a wide range of microstructures, enabling a more objective, robust, and universal characterization of polycrystalline metals.本仓库包含复现我们发表论文中所述结果所需的全部数据与代码。 摘要：从截面多晶金属中识别单个晶粒是微观结构分析的一项基础性任务。然而，传统针对光学显微图像的晶粒分割方法往往面临晶粒间光学对比度不足的问题，且需要人工手动选取可调参数才能获得合格的分割效果。我们提出了一种替代方案，该方案利用了一种被称为定向反射显微术（directional reflectance microscopy）的多角度光学显微技术。通过结合降维、相似-异类分类以及表面定向反射的多区域融合技术，我们的方法可实现多晶表面的全自动且可靠的晶粒分割。我们将该方法应用于具有不同晶体结构与晶粒取向分布的金属样品。实验结果表明，该方法可适用于多种微观结构场景，能够实现更为客观、稳健且通用的多晶金属微观结构表征。