Texture Characterization Using Shape Co-Occurrence Patterns.

Area of Science:

• Computer Vision
• Pattern Recognition
• Image Analysis

Background:

• Texture characterization is crucial for image understanding and pattern recognition.
• Existing shape-based methods capture geometrical aspects but may lack flexibility in analyzing shape relationships.

Purpose of the Study:

• To present a flexible shape-based texture representation using shape co-occurrence patterns.
• To improve texture analysis by incorporating hierarchical shape relationships and high-order statistics.

Main Methods:

• Representing texture images as trees of shapes with associated attributes.
• Learning four types of shape co-occurrence patterns as codewords based on tree hierarchy.
• Investigating three coding methods to encode texture images into descriptive vectors.

Main Results:

• The proposed approach demonstrates robustness to imaging variations.
• It effectively captures geometrical aspects and shape relationships in textures.
• Experiments on diverse datasets confirm the approach's efficiency.

Conclusions:

• Co-occurrence patterns of explicit shapes offer a novel and effective tool for texture analysis.
• This method provides a flexible framework for computing high-order statistics on shape trees.
• The approach advances texture representation in image understanding.