Unsupervised segmentation of carbonate thin section images. (Deliverable D2.1 - Dataset)

Unsupervised segmentation of carbonate thin-section images This data repository is part of Deliverable 2.1 of the Horizon-EU project GO-Forward. It comprises data and processing code for automated mineral phase identification and porosity estimation from thin sections using superpixel segmentation and unsupervised clustering algorithms. The segmentation pipeline uses a two-stage unsupervised approach: 1. SLIC superpixel segmentation: Over-segments each image into ~3000 compact, color-homogeneous regions.2. K-Means clustering in CIELAB color space: Clusters superpixels globally (across all images) into 3 classes based on mean L*, a*, b* and grayscale intensity. Porespace is automatically identified by its distinctive blue-dye signature(negative a* and b* values in CIELAB space). Code is licensed under MIT. Data is licensed under CC BY. More information is provided in the README.md