Automatic extraction of leaf characters from herbarium specimens

Abstract

AbstractHerbarium specimens are a vital resource in botanical taxonomy. Many herbaria are undertaking large‐scale digitization projects to improve access and to preserve delicate specimens, and in doing so are creating large sets of images. Leaf characters are important for describing taxa and distinguishing between them and they can be measured from herbarium specimens. Here, we demonstrate that herbarium images can be analysed using suitable software and that leaf characters can be extracted automatically. We describe a low‐cost digitization process that we use to create a set of 1895 images of Tilia L. specimens, and novel botanical image processing software. The output of the software is a set of leaf characters. As a demonstration of this approach, we extract the length and width of a large number of leaves automatically from images of whole herbarium specimens. We show that the lengths and widths that we extract are very strongly correlated with values in a published account of cultivated species, but are also consistently smaller. We discuss some particular features of herbarium specimens that may affect the results of this form of analysis, and consider further applications to extract characters such as leaf shape and margin characters.