Three Dimensional Segmentation for Cement Microtomography Images using Self-Organizing Map and Neighborhood Features

The performance of cement is strongly influenced by its microstructure, among which dynamic microstructure can reveal the formation and development of the cement paste. Therefore, the investigation of dynamic microstructure enables us to understand the cement hydration and try to improve the cement properties. However, the constituents of cement paste are hard to directly segment by human vision due to the fully mixed phases, a lot of noise and low image definition, which influences phase extraction, substance analysis and the study on the change of material composition. This paper studies the three dimensional image segmentation for cement microtomography images using self-organizing map and neighborhood features. The method takes advantage of the neighborhood features and the fault-tolerance to missing, confusing, noisy data of self-organizing map. The experimental results manifest that this method perform well. Furthermore, the evolution of cement three-dimensional microstructure during hydration is analyzed by the segmented images.