Results
The matching tests succeeded for all the profiles for both species. The program correctly identified all the re-sightings already identified in situ by the presence of flipper tags or a series of photographs taken at the same moment. The system decresed the number of images that needed to be visually compared, to a maximum of 20 images selected from the registered data according to the input profile code. All the query profiles entered for the test matched with at least one of the top six result displayed by the system. And among the matching profiles, 94.44% profiles for green and 78.26% profiles for hawksbills were founf in the first position in the list, which revealed a better accuracy for green turtles. This variation between species appears to be related to the lower number of scutes recorded on the profile for the 14 hawksbill used (N=10.39 SD=1.63) compared to 14 greens (N=17.35 SD=2.57). This resulted in a shorter code for howksbiss made up of fewer 3-digit scute-codes, and consequently to a higher number of potentially matching profiles. For this reason, the entire profile showing all the scutes near the neck is required for hawksbill turtle photo-ID. Conversely, a wide range of photographs can be used for green turtles as long as post-ocular and bottom-central scutes (i.e. at least the two first rows) are visible. Moreover, results showed that blurred photographs could be used, as long as the separations between the scutes were visible.
…