Chord Bunch Walks for Recognizing Naturally Self-Overlapped and Compound Leaves

Area of Science:

• Computer Vision
• Image Analysis
• Botanical Informatics

Background:

• Recognizing leaf shapes from large datasets with variations is challenging.
• Existing methods struggle with arbitrary variations and complex leaf structures.

Purpose of the Study:

• To develop a novel descriptor for robust leaf shape description and recognition.
• To address limitations in current shape analysis techniques for botanical applications.

Main Methods:

• A novel chord bunch walks (CBW) descriptor was developed by measuring chords passing through the shape.
• The CBW descriptor integrates shape image functions, capturing contour and interior properties.
• A Log-Min distance was proposed for efficient and effective CBW matching, ensuring one-to-one correspondences.

Main Results:

• The CBW method demonstrated invariance to rotation, scaling, translation, and mirror transforms.
• Experiments on large datasets showed significant accuracy increases for compound and self-overlapped leaf retrieval.
• The approach achieved high accuracy with low computational costs compared to benchmarks.

Conclusions:

• The proposed CBW descriptor offers a powerful and efficient solution for leaf shape analysis.
• This method has significant potential for applications in botany, ecology, and computer vision.
• The CBW descriptor effectively handles complex leaf shapes and variations, advancing shape recognition technology.