Fin-printing great white sharks Carcharodon carcharias: A fully automated computer vision system for the identification of individual animals

Studies in conservation biology and behavioural ecology rely heavily on data relating to the whereabouts and movement of individual animals, with photo-graphic identification methods providing a non-invasive alternative to capture-mark-recapture protocols. And yet, manual identification schemes require enormous amounts of time and effort on the part of a typically small number of highly trained experts, limiting the density of observations that can be made in this way. In order to overcome this limitation, we aim to exploit state-of the-art computer vision techniques to design systems to automatically identify individual animals. Here, we present results from a prototype system focussing on one of our test-bed species, the great white shark Carcharodon carcharias.
Individual shark identification relies on uniqueness encoded in the jagged morphology of the posterior edge of the sharks’ dorsal fin. This pattern is stable over decades, making it an ideal machine-recognisable biometric identifier. Taking as input high-quality, high-resolution images of a side view of the fin, we first combine fin-contour and species shape models to automatically extract and recognise fin regions, before using points of high curvature to partition the trailing edge contour, and produce a spectral bag-of-words representation of individual identity. Individuals are identified by matching these spectrally represented contour sections against a pre-computed database of known individuals. Using this approach, and on a test dataset of 132 images, we are able to identify the correct individual in 85% of cases, predicting success or failure in 100% of the time.