The Other Side of Facial Recognition

Run the term facial recognition through a search engine, and you will find yourself faced with two different results. Firstly, you get the educational texts and tutorials that introduce you to the concept, inform you of what it means and break down how it works. Secondly, you get the news articles, all detailing its stalker-like potential and how we will never be able to escape the ever-watchful eye. With so much focus on the negative, it is hardly surprising that such a powerful technology has gained itself quite the reputation.

As a break from the formulaic gloom, doom and despondency, we have taken a look at facial recognition from a slightly different perspective. So let us present to you the light at the end of a supposedly dark tunnel in the form of alternative uses for facial recognition.

Wildlife Conservation

The monitoring of animal numbers and behaviour plays a major part in the conservation of endangered wildlife. While traditional methods of tracking and counting, such as tagging and boots-on-the-ground, certainly get the job done, they are far from ideal. Costing quite a pretty penny these methods, often put animals through unnecessary stress or producing inconsistent results.

Now scientists have begun to use facial recognition to observe and track animals from a distance.


Trained from roughly 426 images of 80 red-bellied lemurs, and a mix of 190 images of other lemur species, a proof-of-concept system from the scientists at Michigan State University can identify individual lemurs with a 98% accuracy.

Like most facial recognition systems, key landmark features such as eye spacing and blemishes are used to build an identity. In terms of the red-bellied lemurs, it seems that unique fur patterns are the primary differentiator. With this knowledge in hand, this system may work just as well for species such as raccoons, red pandas and sloths, that all have variable facial hair and skin patterns.

With access to a very limited amount of data, the development team had to manually prepare each image by identifying the locations of eyes and fixing the cropping and orientation to get a suitable portrait. Had they had enough images, a Convolutional Neural Network (CNN) could have done this work automatically. It is hoped that with time and an increase in available training data, the development of a CNN will be possible.


Fig 1  Red Bellied Lemur at the heart of a new feasibility study

Despite the small training set, this system has had some success, however, the biggest issue so far, is that detection is only possible if an animal looks directly into the camera or if it’s fur does not obscure its key features. With enough training, LemurFaceID will become a valuable tool in the collection of data on life expectancy, reproduction rates and the rates of infant and juvenile mortality. There is also hope that via a mobile recognition program, law enforcement will have support in the reporting of lemurs held in captivity as illegal pets.

Chimpanzee Behaviour

As a step up from LemurFaceID, researchers at Oxford have been working to develop a deep learning algorithm to recognise both the sex and identity of chimpanzees.

Fig 2 ⇑  Facial Recognition to study the behaviour of chimps

Unlike LemurFaceID, around 10 million facial images of just 23 chimps were extracted from approximately 50 hours of footage and fed into a deep neural network. The model produced from this data was able to identify animals with up to 93% accuracy, and classify sex correctly up to 96% of the time. When compared to the skills of expert human labellers given only an hour to complete the task, the system performed twice as well and only took a fraction of a second to complete.

When applied to footage of social situations of a group of chimps, the model decided that mothers and infants would spend most of their time together. This finding correctly aligns with known and understood patterns of behaviour, and as such increases confidence in potential future revelations.

Of course, there were still a few inaccurate identifications. In the test against human labellers, the few times the algorithm failed, it had a tendency to mistake a chimpanzees bottom for a face.

Facebook for Whales

The primate kingdom is not the only group of animals benefitting from modern technology. A biologist by the name of Christina Khan had the idea to utilise facial recognition to identify whales.

Unsurprisingly whale “faces” are incredibly similar in appearance. Fortunately, some species, such as the North Atlantic right whale, have distinctive spots on their head that can work much like a fingerprint to identify individuals. Unfortunately, even with this knowledge, the identification process takes up a significant amount of time. Having a system that could recognise unique individuals would free up time for biologists, such as Khan, to study the animals and keep approaching ships out of their path.

In 2016 the whale identification challenge was opened up to see if anyone could develop the desired system. Upon its conclusion, data science company were able to create an algorithm capable of identifying individual whales with around 87% accuracy. Much like the LemurFaceID system, the biggest issue faced by the algorithm was a lack of training data, with only 4,500 images of right whales available for use. If this number were to increase there is hope that accuracy will increase in response.

While this algorithm is limited to the identification of NA right whales, ID software research is being carried out in hopes of identifying humpbacks. Unlike the right whale, humpbacks can be identified by their tail flukes which make for a slightly more complex challenge.

Fig 3 ⇑  Whales may not have distinctive facial features but they can be identified by a facial recognition system

Diagnosing Diseases

Some rare diseases present with symptoms that make it hard for even the most skilled of clinicians to pinpoint the cause, especially in non-European populations. One such disease is 22q11.2 deletion syndrome or DiGeorge syndrome. Affecting from 1 in 3,000 to 1 in 6,000 children, DiGeorge results in multiple defects throughout the body, including a characteristic facial appearance.

Using facial analysis technology, researchers compared a group of 156 individuals, from diverse ethnic backgrounds, confirmed to have the DiGeorge Syndrome to a group of people without. Based on a range of 126 individual facial features, the system made a correct diagnosis for all ethnic groups in the study 96% of the time.

Researchers are hoping that with further advancement, this technology could be expanded so that healthcare providers could send images for analysis, and receive a diagnosis in return. So far this facial recognition system has proven to be accurate in diagnosing cases of Down syndrome, however, the team of researchers behind the DiGeorge study hope it will prove effective in the diagnosis of both Noonan syndrome and Williams syndrome.

Of course, the concerns regarding facial recognition and privacy are justified. After all, there is no data quite as personal to us as our own bodies. We aren’t suggesting that these factors be swept under the rug and ignored, but whether we like it or not this technology is on our doorstep. So maybe it is time to look beyond the Orwellian predictions and accept that there may be a hint of silver outlining the approaching storm clouds.