Three-Dimensional Leaf Edge Reconstruction Combining Two- and Three-Dimensional Approaches

Area of Science:

• Plant Science
• Computational Biology
• Computer Vision

Background:

• Traditional 2D leaf morphology analysis limits understanding of 3D functional aspects.
• Accurate 3D leaf edge reconstruction remains challenging despite advancements in data acquisition.

Purpose of the Study:

• To develop and evaluate a novel method for 3D leaf edge reconstruction.
• To improve the quantitative analysis of plant morphology using 3D data.

Main Methods:

• Combines deep-learning-based 2D instance segmentation for edge detection with structure from motion for camera pose estimation.
• Employs leaf correspondence identification and curve-based 3D reconstruction using B-spline fitting.
• Evaluates performance on virtual and real leaves, considering factors like leaf size and camera noise.

Main Results:

• The method successfully reconstructs 3D leaf edges, particularly for lobed leaves and those with few holes.
• Reconstruction accuracy is challenged by small leaf size, high camera noise, and highly localized variations like serrations.
• Guidelines for setting curve fragment reliability thresholds were developed, with occlusion and leaf size being key factors.

Conclusions:

• This nondestructive 3D leaf edge reconstruction method advances quantitative plant morphology analysis.
• The approach offers a promising avenue for capturing whole-plant architecture through integrated 2D and 3D phenotyping.
• Further refinement is needed for complex, highly variable leaf edge morphologies.