Geodesic Paths for Image Segmentation With Implicit Region-Based Homogeneity Enhancement

Area of Science:

• Computer Vision
• Image Processing
• Computational Mathematics

Background:

• Minimal paths offer optimal solutions for image segmentation and boundary detection.
• Existing methods often struggle with integrating diverse features like anisotropy and region homogeneity.

Purpose of the Study:

• To develop a flexible interactive image segmentation model.
• To enhance boundary detection and image segmentation accuracy by integrating region-based homogeneity.

Main Methods:

• Utilizing the Eikonal partial differential equation (PDE) framework.
• Constructing local geodesic metrics incorporating anisotropic/asymmetric edge features, region homogeneity, and curvature regularization.
• Employing an implicit representation for region-based homogeneity features.

Main Results:

• The proposed model successfully integrates anisotropic and asymmetric edge features with region-based homogeneity.
• A novel method for constructing closed contours from open curves was introduced.
• Experimental results demonstrate superior performance compared to state-of-the-art minimal path segmentation methods.

Conclusions:

• The developed interactive segmentation model offers a flexible and powerful approach.
• The integration of implicit region-based homogeneity and local geodesic metrics significantly improves segmentation quality.
• This work advances minimal path-based image segmentation techniques.