Region-based active contours with cosine fitting energy for image segmentation.

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Traditional active contour models often struggle with image segmentation, particularly in the presence of noise and intensity variations.
• Existing methods may require computationally expensive reinitialization steps for level set functions.

Purpose of the Study:

• To propose a novel active contour model utilizing cosine function-based data fitting for image segmentation.
• To extend the model for handling intensity inhomogeneity and improve computational efficiency.

Main Methods:

• Developed a global active contour model incorporating cosine fitting energy.
• Extended the model to a local cosine fitting energy version to address intensity inhomogeneity.
• Integrated level set regularization to avoid reinitialization costs.

Main Results:

• The proposed global and local cosine fitting energy models demonstrate accurate and effective image segmentation.
• These models exhibit superior efficiency and robustness compared to Chan-Vese and local binary fitting models, especially for noisy images.
• Level set regularization successfully reduced computational overhead.

Conclusions:

• The cosine fitting energy active contour models provide an accurate, efficient, and robust solution for image segmentation.
• The models are particularly advantageous for segmenting challenging images with noise and intensity variations.
• The integration of level set regularization enhances the practical applicability of these segmentation techniques.