Straightening 3-D surface scans of curved natural history specimens for taxonomic research

Two challenges for taxonomists are proper identification of specimens to known species and extracting information from specimens to diagnose new species. Both tasks are complicated by the very large numbers of known and unknown species and the dwindling numbers of qualified taxonomists to identify/diagnose them all. Automated species identification is a tool that can assist taxonomists facing this challenge. This paper looks at one aspect of automated species identification: unfolding curved specimens, which commonly occurs when specimens are prepared for storage in natural history collections. Here we attempt to address the rather extreme case of an elongate fish specimen coiled along its medial axis. The medial axis is the set of all points within an object with the shortest distance to at least two different points on that object’s surface, where “distance” (typically Euclidean) is determined by the application. Medial Axis Estimation is a challenging problem that arises when the surface itself is sampled (i.e. incomplete). In this paper, we look at various techniques for estimating the medial axis of an object, then we propose a new method for medial axis estimation based on localized spatial depth. We extend the idea of localized spatial depth-based medial axis further by applying an original ridge detector. We conclude with a comparison of our approach with The Power Crust approach using artificial data.