Recognition of juvenile hawksbills Eretmochelys imbricata through face scale digitization and automated searching

Abstract

Advancements in digital photography have facilitated the use of photo-ID to track individual animals, making this technique of great value for conservation biology. However, the time required to manually match new photographs to those stored in a database is proportional to the size of the database. Therefore, there is need for investigating the potential to automate the searching processes through computerized means. We encountered hawksbill turtles Eretmo che-lys imbricata (n = 2) that were members of an ongoing study but had lost flipper tags and shell etchings. To identify individuals, we first manually searched photographs of turtles previously captured and released. Manual visual matching of the 2 turtles encountered was successful for 100% of tested photographs. To investigate automated recognition of turtles in a database, we used the spot recognition program, I3S, to digitize scutes on the dorsal and lateral surfaces of the head and to compare spot patterns through the automated system. I3S successfully identified the 2 return turtles as the same turtles identified by the manual visual matching method. To assess the ability of I3S to identify turtles both present in and absent from the database, we blind-tested a series of photographs of turtle heads and faces using both manual visual methods and I3S. With I3S, 84.6% of the computerized photos were successfully matched with photos in the database, with scores produced ranging from 0.069 to 0.435. This study showed the potential for using a photo-database for long-term identification of individual turtles, but that the usefulness of a photodatabase depends on the quality of the photos and the flexibility of the computer program used.