Paper No. 3
Presentation Time: 9:00 AM-6:30 PM


WALKER, Cory L.1, STOUGH, Joshua1, GREER, Lisa2, ELIUM, Elizabeth M.2 and STIER, Arthur F.2, (1)Department of Computer Science, Washington and Lee University, Lexington, VA 24450, (2)Department of Geology, Washington and Lee University, Lexington, VA 24450,

Monitoring changes in coral reefs can be accomplished by taking multiple images of designated areas

of the reef each year and quantitatively comparing live coral abundance from year to year. To determine

percent live coral per m2, we manually segmented all live coral cover in photographs of m2 quadrats

using the pen tool function in Adobe Illustrator. Percent live coral cover was then calculated using

MATLAB. This method is very time-intensive and can take over one hour per image, depending on the

density of living coral. In order to better understand how the coral reefs are changing over time, we

need to speed up this process so that more time can be allotted to collecting and analyzing data. Here

we present a MATLAB program that provides a semi-automatic method for segmenting live coral with

only a few clicks. In our alpha test, we used both the manual and semi-automatic methods on 10 typical

quadrats. For each quadrat we obtain 5 segmentations: two independent manual segmentations, one

collaborative manual segmentation that we consider ground truth, and two independent semi-automatic

results. In evaluating our method, we found that manual segmentations average a 2.5% discrepancy in

estimated live coral coverage, while our method averaged 2.0% discrepancy. Additionally our semi-

automatic method averaged a 5.7% discrepancy with manual results, with a median of 3.15%. We look

to improve our program on two fronts, both to reduce discrepancy with manual raters and to improve

efficiency. With these improvements we can spend more time obtaining data and monitoring changes in

coral reefs.