Contour recognition of complex leaf shapes.

The leaf shape is an important taxonomic character. Compared to the classic morphological leaf features such as veins, margin indentations, sinuses, etc., the shape is simpler to obtain by using the 'magic wand' or other contouring tools that are available in most of imaging applications. The only exception is when leaves develop large lobes that get in touch or overlap each other, as the presence of hidden or closed portions of the leaf border precludes the application of automatic methods and forces the leaf contour to be traced manually. This is a time consuming and relatively accurate operation that, nevertheless, can not be avoided, as overlapping lobes are characteristic features of the leaves of several plant species and varieties. The method described in the paper overcomes this problem as it allows the leaf contour to be achieved even in the presence of touching or overlapping lobes. The method involves three steps: (1) the acquisition of leaf images using a transilluminator, (2) a two-level image segmentation that allows all leaf components (blade, overlapping lobes and closed sinuses) to be represented in a single binary image, and (3) the contouring and concatenation of all binary outlines in a single, self-intersecting closed contour that reproduces accurately the leaf shape. The method can be extended to acquire the shape of leaves of herbarium specimens, that are often overlapped but can not be easily handled and repositioned because of their extreme fragility and relevant taxonomic value.