Did you know that every month the SDRP staff may take more than 5,000 photographs of dolphins in Sarasota Bay during our population monitoring surveys? This results in more than 60,000 (!) images a year that we have to process (rename, crop to fins, etc.) and then match to our photographic catalog of >5,500 individuals for the central west coast of Florida. As you can imagine, this equates to hours upon hours of work for us and can take even longer when we encounter an unfamiliar or potentially new individual. Wouldn’t it be great if we could figure out a way to automate this process? The need for automation is becoming even more pressing as our Gulf-wide collaborative bottlenose dolphin identification catalog, GoMDIS, grows. To date, we have received and incorporated 19 catalogs into GoMDIS, including more than 13,000 dolphins and 23,000 images, with at least one catalog from each U.S. Gulf state as well as Cuba. Based on firm expressions of interest from others, we expect to receive catalogs from many more sites in the near future.
Automatic facial recognition software has been around for a while and has proven to be remarkably successful at correctly identifying people in pictures (have you ever tried to ‘tag’ your friends in an image you post on Facebook and been shocked when it correctly guesses who everyone is?). Automatic detection algorithms for identifying animals such as zebras, whale sharks, manta-rays, and even humpback whales are proving to be very successful and are now being implemented in several large-scale collaborative catalogs to examine the distribution and movements of these species. The one thing that animals such as these listed have in common is some sort of patterning or coloration that the computer algorithms can use to correctly match individuals. Bottlenose dolphins lack these features, making the creation of successful algorithms a bit harder as they must rely almost solely on the nicks, notches and scars of the dolphins’ dorsal fins. To make matters even more difficult, variations in fin angle or fin side (left or right) in an image hinder the ability of these algorithms to successfully match individuals.
Thanks to donations from friends of SDRP (like you!), we raised enough money during this year’s Community Foundation of Sarasota County’s ‘Giving Challenge’ to partner with Duke University and work with WildMe (www.Wildme.org) to support the development of an automatic detection algorithm for bottlenose dolphin fins in Wildbook (http://www.wildbook.org/). The money raised is helping to support a PhD student in the Department of Computer Science at Rensselaer Polytechnic Institute, who will use our extensive photographic catalog of known individuals to develop, test, and refine an algorithm that successfully matches individual dolphins based on their nicks, notches and scars regardless of fin angle or side. Once developed, this program will not only help us more quickly ID Sarasota Bay dolphins, it will help our Gulf of Mexico dolphin research partners identify dolphins in their own regions and support similar dolphin research programs around the world. Thank you to everyone who helped make this endeavor possible!
This article appeared on pages 19-20 of the 2017 SDRP Annual Report, Nicks n Notches.