Computer-assisted recognition of dolphin individuals using dorsal fin pigmentations

Ecologists commonly use photo-identification of individual animals to monitor the behaviour, state and health of a population, since it is a cost-effective technique that eliminates the need to physically capture and tag animals. With dolphins, the nicks and notches of the dorsal fin are typically used as the unique identifying features for each individual; however New Zealand common dolphins are relatively unmarked, so most of the population cannot be identified. Here, we investigate how computer vision can be used to extract information from the pigmentation patterns that are typically seen on adult common dolphin dorsal fins. We develop features that are relatively robust to changes in the fin orientation and compare the classification rates of 779 photos of 169 different adult common dolphins. Using pigmentation-based features, we correctly classified individuals 75% of the time, with our top-5 estimates containing the correct dolphin in 86% of the cases.