An Automatic Hierarchical Leaf Venation Segmentation Based on Region Growing

Leaf venation segmentation plays an important role in understanding the biological significance of leaves and the laws of plant growth. Although the hierarchical venation network can better reflect the characteristics of veins, most of the existing methods only extract the complete vein network without hierarchical segmentation. The few methods available are either semi-automatic that requires manual intervention, or for leaves with smooth edges that do not segment leaves effectively. Thus we proposed a novel venation hierarchical segmentation method based on region growing, which is suitable for segmenting venation of sweet cherry leaves with serrated edges. The performance of the proposed method was evaluated on 80 images of a sweet cherry leaf dataset. The average completeness on the primary vein reaches 0.8658 and on secondary veins reaches 0.6407. It proved the effectiveness of this method, which also had value in leaf phenotype extraction and leaf classification.