Context-aware discovery of visual co-occurrence patterns.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Traditional clustering of visual primitives often overlooks spatial and feature structures.
• This limitation hinders the discovery of complex, high-level visual patterns.

Purpose of the Study:

• To develop a method for discovering meaningful visual patterns by incorporating spatial and feature contexts.
• To improve the accuracy and robustness of visual primitive clustering.

Main Methods:

• Proposed a regularized k-means clustering approach incorporating spatial and feature contexts as constraints.
• Introduced a novel self-learning procedure to iteratively refine clustering results using discovered patterns.

Main Results:

• The proposed method effectively addresses ambiguities in visual primitive clustering.
• Experiments on real images demonstrate the method's superior performance in pattern discovery.

Conclusions:

• Considering spatial and feature contexts significantly enhances visual pattern discovery.
• The self-learning procedure guarantees convergence and validates the method's effectiveness.