Hi,

Unfortunately i can't seem to find the dataset of leaves that they used, otherwise I would have suggested to start by downloading their dataset/toolbox/code, and reproducing their results.  In the face of this problem, If I was you I would probably start by running a standard classification pipeline such as the one in the VLFEAT tutorial, you could always add those features in to see if you better results.  http://www.vlfeat.org/applications/caltech-101-code.html.

Although what you really have here is a fine-grained classification problem, so it may not work so well at tomato leaf a Vs tomato leaf B.  But it is a start at least.

Perhaps you could upload your tomato leaf data-set to get people interested.

Cheers

Hi,

I am not an expert of such problems, yet recent works by Guillaume Cerutti (links attached) may be useful. They focus on leave recognition of European trees and provide generic techniques that have low computational cost (as they are implemented on smart phones).

I hope this can help

Best regards

https://www.researchgate.net/researcher/2007195253_Guillaume_Cerutti

Article Understanding Leaves in Natural Images - A Model-Based Appro...