Random Walker based Automated Phase Discrimination and Morphological Insights in Ultra-High Carbon Steel

Image analysis is an important process required for characterization in different engineering domains. This work presents a novel image processing methodology to segment phases and quantify grain morphology in ultra-high carbon steel (UHCS) micrographs. Unlike conventional global thresholding methods, an interactive random walker technique which leverages user-defined markers for localized phase discrimination was deployed. Adaptive histogram equalization enhances contrast to identify seed markers representing distinct phases. The random walker algorithm propagates labels from these seeds by minimizing an edge-weighted cost function. This interactive approach provides precise phase boundaries despite variability in grain morphology. The segmented output enables statistical grain size, shape, and orientation quantification. The study demonstrates the UHCS micrograph data for an input grayscale image, and the mean area, mean intensity, and mean orientation of all the grains are computed to be 716.667 px
²
, 129.449, and -53.648, respectively. By combining accessible preprocessing, interactive localization, and rigorous postprocessing, our pipeline automates the characterization of microstructural features from image customizable nature of this physics-based machine learning technique makes it widely applicable for robust characterization of varied material microstructures from images.