A Nonlinear Adaptive Level Set for Image Segmentation

Area of Science:

• Computer Vision
• Image Processing
• Computational Mathematics

Background:

• Image segmentation is crucial for analyzing visual data.
• Traditional level set methods (LSM) struggle with weak boundaries and false edges.
• Robust segmentation requires methods that handle image noise and regional features effectively.

Purpose of the Study:

• To develop a novel level set method (LSM) for enhanced image segmentation.
• To improve segmentation accuracy for objects with weak or ill-defined boundaries.
• To create a more robust segmentation technique by incorporating Bayesian principles.

Main Methods:

• A novel level set method (LSM) utilizing the Bayesian rule.
• Design of a nonlinear adaptive velocity for curve evolution.
• Implementation of a probability-weighted stopping force to prevent leakage.
• Involvement of regional image features for automatic curve adjustment.

Main Results:

• The proposed method automatically determines curve evolution direction using Bayesian rules.
• Adaptive speed control prevents leakage at weak boundaries.
• Reduced influence of irrelevant edges and false boundaries.
• Competitive performance demonstrated on artificial, medical, and BSD-300 datasets.

Conclusions:

• The novel level set method offers robust image segmentation, particularly for challenging cases.
• Bayesian rule integration enhances accuracy and adaptability in curve evolution.
• The method provides a competitive alternative to existing level set techniques for various image types.