Simulation of image acquisition in machine vision dedicated to seedling elongation to validate image processing root segmentation algorithms

Abstract

This article proposes a methodology for the numerical validation of image processing algorithms dedicated to the segmentation of roots of plants with machine vision. A simulator of plant growth is coupled to a simulator of the image acquisition to generate images of simulated plants associated with a known synthetic ground truth. The simulator incorporates parameters of the plant and parameters of the experimental imaging system acquiring the images. This opens the possibility to assess the impact of these parameters on the performance of any segmentation algorithm on unlimited populations of virtual plants. Illustrations of this approach are given for the segmentation in 2D of seedlings with several classical algorithms and also with an algorithm of recent introduction. The presented results can be easily extended to 3D and are therefore also appropriate for other segmentation algorithms of roots with imaging modalities adapted for 3D root tracking like X-ray or MRI.