Mutual Learning Between Saliency and Similarity: Image Cosegmentation via Tree Structured Sparsity and Tree Graph...

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Weakly supervised image cosegmentation remains challenging.
• Existing methods often focus on either object saliency or similarity, neglecting their interaction.

Purpose of the Study:

• To propose a unified mutual learning framework for weakly supervised image cosegmentation.
• To integrate structured sparsity and tree-graph matching for improved object delineation.

Main Methods:

• A novel framework based on image hierarchies.
• Integration of structured sparsity with tree-graph matching.
• Exploitation of hierarchically geometrical relationships of coherent objects.

Main Results:

• The framework learns prior knowledge for structured sparsity via tree-graph matching.
• Object-oriented salient regions are generated effectively.
• Reduced searching space and computational complexity for tree-graph matching.

Conclusions:

• The proposed mutual learning framework effectively delineates co-existing object patterns.
• Experimental results validate the framework's superiority over state-of-the-art methods on benchmark datasets.