3D Materials image segmentation by 2D propagation: a graph-cut approach considering homomorphism

Area of Science:

• Materials Science
• Computer Vision
• Image Analysis

Background:

• Segmentation propagation transfers image segmentations across sequences, crucial for 3D reconstruction in materials science.
• Existing methods individually address shape, appearance, or topology, limiting comprehensive segmentation accuracy.

Purpose of the Study:

• To develop an advanced segmentation propagation framework for 3D materials image analysis.
• To improve the accuracy and robustness of segmenting contiguous 3D structures from 2D serial sections.

Main Methods:

• Formulated segmentation propagation as an optimal labeling problem solved via graph-cut.
• Introduced a homomorphic propagation considering region adjacency consistency.
• Integrated shape and appearance consistency into the propagation process.
• Implemented a local non-homomorphism strategy for substructure changes.

Main Results:

• The proposed framework demonstrated effective segmentation propagation on various 3D materials images.
• Experimental results showed superior performance compared to several existing image segmentation methods.
• The framework successfully handled variations in shape, appearance, and topology, including appearing/disappearing substructures.

Conclusions:

• The developed framework offers a significant advancement in materials image segmentation.
• Incorporating multiple structural properties enhances the accuracy of 3D reconstruction from serial sections.
• The graph-cut based approach provides an efficient and effective solution for complex segmentation tasks.