Automating plant phenotyping. We have added an initial approximation to leaf counting and identification. This will detect the leaf centers and give the number of leafs of a plant. Inspiration came from a paper presented at CVPPP2014: 3-D Histogram-Based Segmentation and Leaf Detection for Rosette Plants. Jean-Michel Pape et. al. 2014. And while there is still some false negatives that need to be addressed, I think the use of a simple distance map, together with local maximum is a pretty good starting point. Here is an example image from our data set:
The circles mark the places where we detect a leaf. Notice how the plants in the right did not get detected, this is due to segmentation errors.
The circles mark the detected leaf centers. Some of the leafs of these plants are not detected because they don’t translate into a detectable maximum in the distance transformation.