Exploiting species-distinctive visual cues towards the automated photo-identification of the Risso's dolphin Grampus griseus

The photo-identification is largely employed technique used by biologists in numerous studies, based on a noninvasive approach and aimed to the identification of an individual starting from multiple images. The procedure is based on the exploitation of discriminating features and on the fundamental hypothesis that a single individual can be uniquely recognized if depicted on an image. Currently, this technique is effectively used to investigate on spatial/temporal wild species distributions or, generally speaking, to improve knowledge on data-deficient species. In this paper we focus on an innovative computer vision approach, aimed to the automatic photo-identification of Risso's dolphins, based on Speeded Up Robust Features (SURF) computed on the dorsal fin to recognize an unknown individual among a set of models with a best matching approach. Experiments on real data acquired in the Gulf of Taranto as well as a comparison with the state-of-the-art DARWIN software confirm the profitability of the proposed approach in terms of accuracy improvements and reduced computational time.