Sea Turtle Facial Recognition Using Map Graphs of Scales

Abstract

AbstractIndividual identification of sea turtles is important to study their biology and aide in conservation efforts. Traditional methods for identifying sea turtles that rely on physical or GPS tags can be expensive, and difficult to implement. Alternatively, the scale structure on the side of a turtle’s head has been shown to be specific to the individual and stable over its lifetime, and therefore can be used as the individual’s “fingerprint”.Here we propose a novel facial recognition method where an image of a sea turtle is converted into a graph (network) with nodes representing scales, and edges connecting two scales that share a border. The topology of the graph is used to differentiate species.We additionally develop a robust metric to compare turtles based on a correspondence between nodes generated by a coherent point drift algorithm and computing a graph edit distance to identify individual turtles with over 94% accuracy.By representing the special and topological features of sea turtle scales as a graph, we perform more accurate individual identification which is robust under different imaging conditions and may be adapted for a wider number of species.