Edge grouping combining boundary and region information

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• Image noise significantly degrades the detection of salient structures.
• Existing edge-grouping methods often struggle with noise and may miss larger structures.

Purpose of the Study:

• To introduce a new edge-grouping method for detecting perceptually salient structures in noisy images.
• To improve the robustness of edge grouping against image noise.

Main Methods:

• A novel grouping cost function is defined in a ratio form: boundary proximity over structure area.
• A specialized graph model with two edge types is developed to represent the grouping problem.
• The problem is reduced to finding a minimum-cost cycle in the graph, solvable via a known graph algorithm.

Main Results:

• The proposed method demonstrates increased robustness to image noise due to the area term in the cost function.
• Experimental results on synthetic and real images show competitive or superior performance compared to existing methods.
• The method successfully detects larger-size structures, a key advantage in noisy environments.

Conclusions:

• The new edge-grouping method offers a robust solution for salient structure detection in noisy images.
• The graph-based approach provides an efficient and effective way to find optimal edge groupings.
• Extensions incorporating continuity, intensity homogeneity, and boundary balancing offer further improvements.