[Hierarchical image segmentation based on watershed filtering and fuzzy cluster].

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Context:

• Watershed algorithms are widely used for image segmentation, generating closed outlines.
• A common limitation is over-segmentation, resulting in numerous small regions that obscure target shapes.
• Existing methods often struggle to effectively merge these small regions.

Purpose:

• To propose a novel scheme for merging small regions generated by watershed algorithms.
• To develop a hierarchical segmentation representation that mitigates over-segmentation.
• To introduce a dissimilarity measure for neighboring regions and utilize fuzzy matrix calculations.

Summary:

• The paper presents a method to merge oversegmented regions from watershed algorithms.
• A dissimilarity measure between adjacent regions is defined.
• Fuzzy matrix calculation and hierarchical segmentation are employed to refine the segmentation results.
• Experiments demonstrate the effectiveness across various image types.

Impact:

• Improved accuracy and clarity in image segmentation, particularly for complex shapes.
• Provides a robust hierarchical representation for image analysis.
• Offers a solution to a significant challenge in automated image segmentation.