Coral reefs are on the brink of collapse throughout the Caribbean, and there is a great need for an automated method to assess the health of the reefs more quickly, according to Lisa Greer, associate professor of geology at Washington and Lee University.
So when she heard about the computer vision research of her W&L colleague Joshua Stough, assistant professor of computer science, she invited him to collaborate on producing such a method.
"The kind of work we are doing is a perfect fit for W&L," said Greer. "At a small school you are more likely to hear about what faculty in other departments are doing and I heard about Joshua's work from a colleague in the biology department."
For his part, Stough wasn't sure that his expertise was a good match for working with coral reefs, at least not when Greer first approached him.
But he decided to give it a try, and last summer the two W&L scientists co-authored a paper on their research which was presented at the International Coral Reef Symposium in Australia. Their paper, "Texture and color distribution-based classification for live coral detection," was co-authored with Matt Benson, a 2012 graduate from New Orleans, and William Sullivan, a sophomore from Vienna, Va.
Greer described the current method of assessing the health of coral reefs as "incredibly painstaking. We literally trace digitally on a computer the live coral versus the dead coral. This new machine learning technique will be able to accurately recreate that work in a fraction of the time it takes us to do it manually. If this works, then people throughout the coral reef community could use it to create rapid assessments of this type of coral reef."
Greer and her students are tracing the live coral in a particular species of endangered coral that is found in abundance in only a very few places, such as the north coast of the Dominican Republic and Roatan, Honduras. The coral forms thickets that look like branches piled up together so it is possible to see the skeleton of the coral forming in the thickets. However, it is hard to assess how much is live coral as opposed to a dead skeleton covered by algae.
"People in coral reef studies have argued for years about what a healthy coral reef looks like and how much live coral is necessary to call a reef healthy," said Greer. "When you're measuring the health of something, you have to compare it to what it was before or to a similar system and it must be quantitative."
Stough's computer vision technique essentially trains the computer to recognize live coral by looking at its color and texture. He explained that much computer vision research can be summarized as considering various dimensions of the image data — how much blue is in one area, or how blurry or pronounced the edges are — then using machine learning, or prior knowledge, to determine patterns in those dimensions that allow people to discriminate or make conclusions about the images.
"With coral reefs the dimensions are different, such as the local distribution of color in the image, and the question is different, but the tools brought to bear for determining the patterns are the same," said Stough.
Initially, Stough and Greer worked on a test case using several underwater photographs of coral reefs Greer and her students had collected off the coast of Belize. "We achieved strong results in the test case, and we are past phase one of trying to see the utility in this," said Greer. They plan to add a second data set of 120 photographs during the summer of 2013.
"When we started, I knew nothing about the kind of work Joshua does, and he knew nothing about coral reefs," she added. "We've had a lot of fun learning about each other's fields. It's so nice to be working with someone outside of geology."
Stough pointed out that computer scientists often do collaborative work. "One focus of computer science is the computer scientist as a 'toolsmith,' where we work directly with domain experts like Lisa and help them use computers to solve the harder, or more time-consuming, problems in their field," he said.
"Improving rapid assessment techniques is of course just one way to contribute to the science focused on reversing coral reef decline. Attending the conference in Australia with Lisa was great. I got to meet so many people passionate about their work with coral reefs, with different perspectives from fields like chemistry, ecology and biology. I am definitely motivated to see more coral reefs before they change too drastically."