Local or global minima: flexible dual-front active contours

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Variational active contour models often rely on local minima, which can lead to undesirable segmentations due to noise or complex image structures.
• Region-based energy functionals offer robustness but are limited to specific image types, unlike more versatile edge-based methods.

Purpose of the Study:

• To develop a novel active contour implementation that overcomes the limitations of existing models regarding local minima and global assumptions.
• To provide a flexible method for image segmentation that can achieve desirable minima, balancing local and global characteristics.

Main Methods:

• A dual front implementation of active contours is proposed, inspired by minimal path techniques.
• Fast sweeping algorithms are utilized to efficiently compute the active contour evolution.
• The model allows manipulation of active region sizes to control the degree of localness and globalness.

Main Results:

• The proposed model demonstrates the ability to achieve minima with variable degrees of localness and globalness.
• Experiments on 2D and 3D images show improved segmentation performance compared to existing active contour and region-growing methods.
• The flexibility allows for obtaining desirable segmentations that are neither purely local nor purely global.

Conclusions:

• The novel dual front active contour model offers a fast, flexible, and effective approach to image segmentation.
• This method enhances the ability to find desirable minima by controlling the balance between local and global search.
• The model's adaptability makes it suitable for a wide range of segmentation applications across different image types.