Measuring and analyzing color and texture information in anatomical leaf cross sections: an approach using computer vision to aid plant species identification

Abstract

Currently, studies on leaf anatomy have provided an important source of characters helping taxonomic, systematic, and phylogenetic studies. These studies strongly rely on measurements of characters (such as tissue thickness) and qualitative information (structures description, presence–absence of structures). In this work, we provide a new computational approach that semiautomates the collection of some quantitative data (cuticle, adaxial epidermis, and total leaf thickness) and accesses a new source of information in leaf cross-section images: the texture and the color of leaf tissues. Our aim was to evaluate this information for plant identification purposes. We successfully tested our system identifying eight species from different phylogenetic positions in the angiosperm phylogeny from the neotropical savanna of central Brazil. The proposed system checks the potential of identifying the species for each extracted measure using the Jeffrey–Matusita distance and composes a feature vector with the most important metrics. A linear discriminant analysis with leave-one-out to classify the samples was used. The experiments achieved a 100% success rate in terms of identifying the studied species accessing the above-described parameters, demonstrating that our computational approach can be a helpful tool for anatomical studies, especially ones devoted to plant identification and systematic studies.