Identification of scanned leaf specimens from tropical forest plants

Title: Identification of scanned leaf specimens from tropical forest plants using venation
pattern and lamina outline.
Supervisor: W.D. Hawthorne, Dept. Plant Sciences
Email [email protected]
Objective. Image processing of high resolution scanned images of leaves from West African
tropical trees, to yield metrics for leaf venation and lamina (leaf or leaflet) outlines that can
then be used to search image databases, helping identification of unidentified specimens.
Project description.
We collect many thousands of specimens per year of plants from tropical vegetation
identification, to help identify species for various localities, ultimately helping to conserve
biodiversity. Many thousands of specimens have already been scanned on an A3 scanner at
600 to 1200 dpi. Our databases include links to these images and other images, and the
metadata about the vegetation where the plants were collected. Identification of new
specimens involves comparing unidentified specimens to those already named. The process
would be facilitated if our database held searchable, and automatically generated,
parameters defining the outlines and venation patterns. The student would research the
existing literature and experiment with software and statistical tools to, for instance, 1.
Extract vector from the raster imagery showing leaf outline and venation patterns, the
student scanning new specimens where required; 2 experiment with parameterisation of
these patterns in a way which highlights species-specific traits; 3. Test the results to try to
identify new specimens of the same species, statistically. They would focus on a limited
subset of species. (If the method showed promise, we would ultimately aim to improve the
efficiency in a future D.Phil. project and build a website enabling accurate identification of
scanned leaves of many tens of thousands of tropical species.)
Prequisities.
The student should already have a strong interest in, and some experience of image
processing and software for pattern recognition. They should be very computer literate and
fluent in R or equivalent. No botanical experience required, but an interest would be useful.
Note the supervisor will be able to offer almost no pattern analysis skills, but will be able to
advise on botany, databases etc.