Leaf Segmentation and Tracking in Arabidopsis thaliana Combined to an Organ-Scale Plant Model for Genotypic Differentiation

Images of Arabidopsis Thaliana automatically acquired by a high-throughput platform such as the Phenoscope contain valuable information on the plant phenotype. A novel image processing algorithm involving both segmentation and tracking of the plant leaves allows to extract areas of the latter. First, all the images in the series are segmented independently using a watershed-based approach. A second step based on ellipsoid-shaped leaves is then applied on the segments found to refine the segmentation. Taking into account all the segments at every time, the whole history of each leaf is reconstructed by choosing recursively through time the most probable segment achieving the best score, computed using some characteristics of the segment such as the orientation, the distance to the plant mass center and the area. These results are compared to manually extracted segments, showing a very good accordance, with numerous and low-biased data for leaf areas. Such data can therefore be exploited to design an organ-scale plant model, adapted from the existing GreenLab model for Arabidopsis Thaliana, and subsequently parameterize it. This calibration of the model parameters should pave the way for differentiation between the Arabidopsis genotypes.