Biometry: the characterisation of chestnut-tree leaves using computer vision

The Department of Biology of the University of Tras-os-Montes e Alto Douro analyses every year a large number of chestnut-tree leaves, in order to measure their biometric characteristics, namely the leaf area, dimensions of the enclosing rectangle, number of teeth and secondary veins. Because for a human operator this is a time consuming and error prone task, a computer vision system has been set up to improve the process. The task of measuring the leaf presents no major problems, while counting the number of teeth and secondary veins has proved to be complex at the resolutions used. This paper describes the state of the project, going into some detail on the algorithms. A complete system has been assembled, based on a PC connected to an imaging system. A windows-based application has been developed, which integrates the control of the operations to grab, store and analyse images of different varieties of chestnut-tree leaves in an organised way. Because the accuracy of the computer vision algorithms used is not sufficient for the system to be completely autonomous, a semi-automatic solution has been adopted. The operator validates or corrects the results of the automatic analysis. This solution leads to a significant improvement in the performance of the human operator, both in terms of speed and quality of the results.