Fig. 4

Steps of defining global leaf ROI. a–d A number of experiment-related problems encountered whilst developing Leaf-GP (to the left of the figures) and results of our solutions (to the right of figures). e A pseudo vegetative greenness image generated. f, g Using k-means to estimate how many colour groups can be classified from a given image. h The classification result of the k-means approach based on the pseudo vegetative greenness picture, highlighting green pixels in red. i A global adaptive Otsu thresholding used to generate a global leaf ROI binary mask. j, k A LAB colour space approach used to extract leaf ROI objects at the image level to improve the global leaf ROI result