Computer-aided photograph matching in studies using individual identification: An example from Serengeti cheetahs

Abstract

Photographic identification of naturally marked animals is a powerful and nonintrusive technique for obtaining information on behavior, population size, and life-history parameters in wild populations. Yet handling large quantities of photographs is time consuming and prone to error. Computer-aided matching can limit the number of photographs that must be examined visually to confirm that 2 sightings are the same individual. To identify individuals, I used a 3-dimensional (3-D) computer-matching system to aid in matching nearly 10,000 photographs of Serengeti cheetahs, Acinonyx jubatus, taken over 25 years. Accuracy in matching 2 photographs increased to 100% as the computer-generated similarity coefficient increased to 0.600 on a scale from 0 to 1. Probability of missing a match decreased to 6.4% when I used a threshold of similarity of 0.450. Poor quality of photographs decreased accuracy and resulted in up to 33% of matches being missed. Comparisons of photographs at skewed camera angles generally reduced similarity coefficients. Similarity coefficients were no higher for related or unrelated animals, suggesting that the technique is not appropriate for distinguishing subtle similarities. Because 3-D computer-aided matching does not require familiarity with distinctive features of the particular study animal, it is robust to matcher inexperience. This technique can be modified for other species that have complex and variable pelage patterns.