The Dark Secrets of Facial Recognition

As the world is advancing, the area of technology is also entering into a new dimension allowing people to access the information they need in no time. One of the main examples of such advancements we have noticed in the last couple of years is the face recognition technology. Biometric innovation has now become a race, various projects are entering the technology market to compete with one another. Software giants are regularly publishing their groundbreaking theoretical discoveries with the focus on image recognition, artificial intelligence, and face analysis to add in our understanding. Face recognition service is a common approach in the area of law enforcement, such as airports and police stations. The biometric technology can unlock your smartphone or you just have to look into the camera even for making a bank transaction. 

“Three technical fields are said to be the vital contributors to the spread of facial recognition service, which are deep convolutional neural networks, big amounts of data, and power GPUs (graphic processing units).”

“One of the challenges for face recognition for today is large volumes of data. Most open-source datasets are not suitable for commercial projects. It’s hard and expensive to collect data and create datаsets based on it. For face recognition projects, you need to use facial data, which is sensitive. And it means that you need to get permission to use it.”

Andrei Alkhouski, Machine Learning R&D Engineer from InData Labs, a face recognition service provider. 

In the real world, the algorithm of facial recognition has been plagued by bias. The problem emerged due to neural networks when they are programmed with different numbers of facial patterns from various people. So, if a system is programmed to identify faces of white males, but fewer people of color or females, the accuracy of the system will be compromised for the general population. Ineffective accuracy can be viewed as misidentification with the possibility of stopping and questioning people wrongfully. The American Civil Liberties Union (ACLU) last year highlighted an unusual function of Amazon Rekognition Software. 28 Congress members were identified wrongfully as individuals with previous records of arrests. The misidentification included Latinos and African-Americans, whereas the company claimed that incorrect settings were used by the ACLU. 

Further facial recognition service shortcomings were put forth by Police trials. A review presented by Cardiff University in South Wales found that the NeoFace system developed by NEC freezes, crashes, and lags when a large number of people are needed to be identified. Not only this, the worse performance of the system affected the workflow, especially during rush hours, such as late afternoons. The reason for this was ramped up cameras due to light sensitivity causing distorted footage. The mugshot database maintained with the help of facial recognition service mostly based on people of color, which as mentioned above, is a biased practice of policing, said by the civil rights groups. 

People can be found through the technology, even if they were arrested before wrongly for a crime they did not commit. The use of facial recognition service by law enforcement is viewed as controversial. Facial recognition technology is argued as a factor responsible for exacerbating “existing and historical bias” to damage a community that is already under the radar of police for surveillance. Meanwhile, technology firms are investing efforts in their system improvement so that they could work faster, recognizing more images, even the ones with bad quality or light.