A Generalized Asymmetric Dual-Front Model for Active Contours and Image Segmentation.

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• Voronoi diagram-based dual-front schemes are effective for image segmentation and domain partitioning.
• Existing models rely on geodesic metrics for distance estimation and Voronoi diagram generation.
• A key challenge is preventing segmentation contours from getting stuck in suboptimal positions.

Purpose of the Study:

• To introduce a new dual-front image segmentation model utilizing asymmetric quadratic metrics.
• To enhance the robustness of image segmentation, particularly for complex image features and distant initial contours.
• To integrate the model with various region-based homogeneity terms for versatile application.

Main Methods:

• Developed a dual-front model incorporating asymmetric quadratic metrics.
• Integrated image features with a vector field derived from the evolving contour.
• Employed asymmetry enhancement to improve contour evolution and prevent false local minima.

Main Results:

• The proposed model demonstrated reduced risk of contours getting stuck at false positions.
• Achieved encouraging segmentation results on both synthetic and real-world images.
• Showcased applicability with different region-based homogeneity terms.

Conclusions:

• The novel dual-front model with asymmetric quadratic metrics offers improved performance in image segmentation.
• The asymmetry enhancement is crucial for handling challenging image characteristics and initial conditions.
• The model provides a flexible and effective tool for advanced image segmentation tasks.