Image denoising with dominant sets by a coalitional game approach.

Area of Science:

• Graph theory
• Computational intelligence
• Data mining

Background:

• Dominant sets offer a new approach to graph partitioning and data clustering.
• Existing methods may lack stability or computational efficiency.
• Coalitional game theory provides a framework for modeling cooperative behaviors in data points.

Purpose of the Study:

• To adapt dominant sets for data clustering using a coalitional game model.
• To introduce 'betrayal' and 'hermit' rules for stable graph partitioning.
• To develop an approximate algorithm for dominant set identification and apply it to image denoising.

Main Methods:

• Formulated dominant set problem within a coalitional game framework.
• Defined 'betrayal' and 'hermit' rules to guide player cooperation and group formation.
• Designed an approximate algorithm for efficiently finding dominant sets.
• Applied the algorithm to image denoising by treating pixels as players seeking similar neighbors.

Main Results:

• Achieved optimal and stable graph partitions through cooperative game rules.
• Developed a computationally feasible approximate algorithm for dominant set detection.
• Significantly improved nonlocal means image denoising by leveraging dominant sets.
• Restored intrinsic image structures and yielded competitive denoising results.

Conclusions:

• The coalitional game model effectively addresses dominant set partitioning.
• The proposed approximate algorithm is efficient and applicable to real-world problems like image denoising.
• Dominant set partitioning enhances image denoising by averaging effects within similar pixel groups.