RESLS: Region and Edge Synergetic Level Set Framework for Image Segmentation.

Area of Science:

• Computer Vision
• Medical Imaging
• Image Processing

Background:

• Active contour models with level set evolution are widely used for image segmentation.
• Hybrid models combining region and edge information improve performance but lack theoretical foundation for term collaboration.
• Existing hybrid models face challenges in weight coefficient selection and multi-modality accommodation.

Purpose of the Study:

• To propose a novel Region and Edge Synergetic Level Set (RELS) framework.
• To establish a theoretical foundation for the collaboration between region and edge information in image segmentation.
• To develop a method for reliably selecting energy term weights and improving segmentation performance.

Main Methods:

• Developed the RESLS framework incorporating a normalized intensity indicator function.
• Embedded region information into an edge-based model for seamless integration.
• Deduced a global optimization condition to constrain energy weights.
• Demonstrated generality using representative and state-of-the-art models.

Main Results:

• The RESLS framework enables reliable selection of weighting parameters guided by the optimization condition.
• Segmentation accuracy, robustness, and computational efficiency were improved compared to component models.
• The normalized intensity indicator function facilitates the embedding of region information into edge-based models.

Conclusions:

• RELS provides a robust and theoretically grounded approach for hybrid level set image segmentation.
• The proposed framework offers improved performance and adaptability across different imaging modalities.
• This method addresses key limitations in existing hybrid active contour models.