Local optimization based segmentation of spatially-recurring, multi-region objects with part configuration...

Area of Science:

• Computer Vision
• Medical Image Analysis
• Computational Geometry

Background:

• Image segmentation accuracy is improved by incorporating prior knowledge.
• Region containment and exclusion are powerful geometric constraints for segmentation.
• Level set methods naturally handle topological changes in evolving contours.

Purpose of the Study:

• To augment the level set framework with region containment and exclusion constraints.
• To enable segmentation of spatially-recurring objects with geometric constraints.
• To develop a computationally efficient segmentation method for large-scale biomedical images.

Main Methods:

• Augmented level set framework incorporating containment and exclusion constraints.
• Inclusion of a distance constraint between boundaries of multi-region objects.
• Validation on biomedical applications and comparison with discrete domain methods.

Main Results:

• Successful segmentation of spatially-recurring objects while satisfying geometric constraints.
• Demonstrated utility and advantages of the augmented level set framework, even with rough initialization.
• Analysis of trade-offs in metrication error, memory usage, and runtime compared to discrete methods.

Conclusions:

• The augmented level set framework effectively incorporates geometric prior knowledge for image segmentation.
• The method provides a computationally efficient alternative for segmenting complex biological images.
• The framework offers practical advantages in memory usage and runtime without significant loss of segmentation quality.