2.2 Image classification
We used the software ilastik [9] to classify all image pixels as either root hair, root or background. The software provides a user interface to create training sets by manually labelling pixels according to their class and to look interactively at the results. For the classification, we used a random forest classifier with features obtained from morphological filters (validation dataset: σ=0.3-10; demonstration dataset: 0.3-50). In the validation dataset, each image was trained and classified separately, while for the demonstration dataset we classified each genotype separately using ten to eleven images from the corresponding genotype.