Enhanced Decoupled Active Contour Using Structural and Textural Variation Energy Functionals.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Active contours are widely used for object segmentation, but face challenges with parameter sensitivity and computational complexity.
• Decoupled active contour (DAC) models improve efficiency by separating internal and external energy optimization.
• Existing DAC methods struggle with weak object edges due to reliance on edge gradients for external energy.

Purpose of the Study:

• To develop an enhanced decoupled active contour (EDAC) model for more robust object boundary detection.
• To address the convergence issues of DAC models in images with weak object boundaries.
• To incorporate texture information explicitly into the active contour segmentation process.

Main Methods:

• A sparse texture model was developed to explicitly consider texture for boundary detection.
• External energy was redefined as a weighted combination of textural and structural variation maps.
• A multifunctional hidden Markov model was employed for enhanced object boundary detection.

Main Results:

• The enhanced DAC (EDAC) model demonstrated effective object boundary extraction by combining texture and structural information.
• Qualitative and visual analyses on natural and Brodatz image datasets confirmed EDAC's performance.
• The approach achieved robust boundary detection without increasing computation time or requiring color information.

Conclusions:

• EDAC offers a significant improvement over traditional active contour methods, particularly in challenging image conditions.
• The integration of texture and structural variations enhances the accuracy and reliability of object segmentation.
• This method provides a computationally efficient and versatile solution for image segmentation tasks.