Group average difference: a termination criterion for active contour.

Area of Science:

• Image processing and computer vision
• Computational anatomy
• Medical image analysis

Background:

• Active contour models (snakes) are widely used for image segmentation.
• Existing methods often require manual parameter tuning or energy functional modification for termination.
• A robust and automatic termination criterion is needed for efficient segmentation.

Purpose of the Study:

• To introduce a novel termination criterion for active contour models based on area difference.
• To evaluate the performance of this criterion in terms of speed, accuracy, and robustness.
• To demonstrate its applicability in automatic segmentation of various shapes.

Main Methods:

• Developed a termination criterion based on the fluctuation of contour area difference during evolution.
• Implemented and tested the criterion with parametric gradient vector flow active contours.
• Incorporated contour resampling and normal force selection for comprehensive evaluation.

Main Results:

• The proposed metric showed a steadily decreasing trend, indicating convergence.
• Achieved significant total time reduction (approx. 50-60%) compared to pixel movement-based methods.
• Demonstrated comparable accuracy to existing methods on synthetic and real medical images.
• Exhibited insensitivity to shape variations and contour resampling.

Conclusions:

• The area difference-based termination criterion offers an efficient and robust alternative for active contour segmentation.
• It enables automatic segmentation with reduced computation time and consistent accuracy.
• The criterion shows potential for broader application in various active contour models (snakes).