A Laplacian-based grayscale image segmentation algorithm for multi-modal pore characterization of geomaterials

Threshold segmentation is a key image processing technique that leverages grayscale information to extract regions of interest, such as pores in geomaterials. Images with bimodal or multi-modal grayscale histograms typically allow effective thresholding, but geomaterial images often exhibit unimodal histograms with a broad central peak and subtle transitions, posing challenges for accurate pore-matrix segmentation.This study proposes an improved threshold segmentation method tailored for such images. The algorithm employs Retinex enhancement to improve contrast, followed by Laplacian absolute value convolution to enhance edge features, reconstructing the histogram to narrow the threshold search interval. The optimal threshold is determined using maximum between-class variance. The proposed algorithm is validated using a variety of imaging modalities, including SEM, FIB/SEM, and X-ray CT. Its performance is compared with commonly used segmentation algorithms and evaluated against MIP (Mercury Intrusion Porosimetry) results at the same scale. The results confirm the robustness, efficiency, and cross-modality applicability of the method, making it a reliable preprocessing step for digital rock analysis and pore-scale simulations in geomechanics and petroleum engineering.