An Efficient Two-Stage Level Set Segmentation Framework for Overlapping Plant Leaf Image
In this paper, an efficient two-stage segmentation framework was proposed to address the plant leaf image with overlapping phenomenon, which is built based on the leaf approximate symmetry and level set evolution theory. In the pre-segmentation stage, a straight line was manually set on the target leaf to approximate the principal leaf vein and the Local Chan-Vese (LCV) model was used on the global image region to help searching the so-called un-overlapping contour in target leaf. In the formal segmentation stage, the symmetry detection was performed based on the pre-defined approximated principal vein to obtain the narrow-band evolution region and the second initial contour. Next, the LCV model was once again used to find the complete target leaf contour in the narrow-band evolution region. Finally, experiments on some real leaf images with overlapping phenomenon have demonstrated the efficiency and robustness of the proposed segmentation framework.
Keywordsapproximate symmetry level set overlapping leaf principal vein two-stage segmentation
Unable to display preview. Download preview PDF.
- 8.Li, C.M., Xu, C.Y., Gui, C.F., Fox, M.D.: Level Set Formulation without Re-Initialization: A New Variational Formulation. In: Proc. IEEE International Conference on Computer Vision and Pattern Recognition (CVPR 2005), pp.430–436 (2005)Google Scholar