In the last century, we lost many of our magnificent animal species including the Honshu wolf, California grizzly bear, Tasmanian tiger, Barbary lion, Caribbean monk seal, Arabian ostrich, and Japanese sea lion. Additionally, many other species are facing the risk of extinction. Among them are the lemurs – you might remember them as the cute fuzzy creatures from the movie Madagascar. Lemurs are a unique group of primate endemic to Madagascar Island in Africa and are considered to be the most threatened mammalian species on Earth. According to the International Union for Conservation of Nature (IUCN) red list of threatened species, out of the 111 lemur species, 24 are critically endangered, 49 are endangered and 20 are vulnerable. This highlights the urgent need to develop conservation strategies for these animals.
In order to do so, it is important to acquire knowledge of behavior, ecology, and evolution of various lemur species, including data on life history, fitness, longevity, and reproductive patterns. Such data can be acquired through long-term studies of known sets of lemurs. However, long-term studies are limited by the difficulties in tracking the known individuals over extended periods of time. The most commonly used method of lemur identification is by capturing and tagging them with unique identifiers. However, this method is expensive, can cause harm to the animals and is not suitable for large scale studies. Alternatively, the researchers rely on the variations in the appearances of lemurs, such as the differences in body size and shape, to identify them. But this is highly subjective and prone to errors and also requires substantial training of the researchers. Addressing these problems, scientists (Crouse et al.) recently published a study in BMC Zoology, where they modified the human facial recognition technology to develop a highly accurate computer-assisted lemur facial recognition system termed as LemurFaceID. This system uses the variations in the facial patterns of the lemurs for their identification based on the photographs.
For the prototype development, the researchers generated a dataset of 462 photographs of 80 red-bellied lemurs (Eulemur rubriventer) mostly from the individuals in Madagascar. Additionally, to increase the size of the lemur photo gallery, another database was generated that contained the images of lemurs belonging to other species. Each image in the database was subjected to multiple pre-processing steps and further normalizations were performed to reduce the effects of the ambient illumination and lemur’s facial hair on the accuracy of LemurFaceID. The corrected image was subjected to feature extraction using multi-scale local binary pattern (MLBP) method. The final feature vector was constructed based on the linear discriminant analysis (LDA), which helped to minimize the variations between the photographs of the same individual. To perform the face matching, the lemur dataset was divided into (i) a training set which was used to train the LemurFaceID system and (ii) a testing set which was used to test the accuracy of this system. Further, in the test set, two-thirds of the images of each individual were used as a gallery in the system database, while the remaining one-third of the images were used as queries. Each query consisted of one or more images which were identified against the gallery database.
The researchers conducted the face recognition experiments in two different modes. The open-set mode was based on the assumption that during the experiments, queries might be encountered that may not match with any of the images in the gallery. This corresponds to the conditions in the wild, where one might encounter novel lemur individuals which were not spotted before and are consequently absent from the dataset. On the other hand, experiments in the closed-set mode were performed with the assumption that all the query lemurs were present in the gallery. This simulates the condition in the captive lemur colonies where all the individuals are already identified. Across a 100 trials performed in the closed-set mode, LemurFaceID identified lemurs with an accuracy of about 93.3% for a 1-image query and 98.7% for a 2-image query. However, the results with the open-set mode were less accurate suggesting a need to further improve the technique perhaps by increasing the size of the lemur database. In the future, the researchers plan to test the system in the field to compare its accuracy with that of the trained and untrained field observers.
The LemurFaceID provides a novel tool that will greatly facilitate the long-term research of known lemur populations and will help to develop informed strategies for lemur conservation. As lemurs also face the threat of being live-captured to be kept as pets, this technique can be developed into a tool to identify the captive lemurs and report their sightings. IUCN has started the lemur conservation program under the auspices of Save Our Species (SOS) initiative and has been trying to tackle various threats faced by lemurs. LemurFaceID can boost the IUCN’s efforts to conserve the lemur populations. In the future, face recognition tools similar to LemurFaceID can be developed for other animals that show similar variations in facial and skin patterns, such as bears and red pandas. Such innovative approaches, combined with advanced technology, have the potential to create better solutions for conserving our biodiversity.
Crouse D, Jacobs RL, Richardson Z, Klum S, Jain A, Baden AL, Tecot SR. LemurFaceID: a face recognition system to facilitate individual identification of lemurs. BMC Zoology. 2017, 2:2. DOI: 10.1186/s40850-016-0011-9.
Featured image source: Pixabay
About the author:
Isha Verma is currently pursuing her PhD in Stem cell research from the Indian Institute of Science, Bangalore. She loves reading and traveling.
Edited by: Radhika Raheja