Automated Pollen Image Classification

This Master of Science thesis reviews previous research, proposes a method anddemonstrates proof-of-concept software for the automated matching of pollen grainimages to satisfy degree requirements at the University of Tennessee. An ideal imagesegmentation algorithm and shape representation data structure is selected, alongwith a multi-phase shape matching system. The system is shown to be invariantto synthetic image translation, rotation, and to a lesser extent global contrast andintensity changes. The proof-of-concept software is used to demonstrate how pollengrains can be matched to images of other pollen grains, stored in a database, thatshare similar features with up to a 75% accuracy rate.