Law Enforcement Agency of India are starting to take facial recognition technology seriously. Delhi Police, looking into identification people involved in civil unrest in northern India over the past few years, they said they would consider an accuracy of 80% or more a “positive” result, according to documents obtained by the Internet Freedom Foundation through a records request public.
The arrival of facial recognition in India’s capital region marks expansion of Indian law enforcement officials using facial recognition data evidence of possible prosecution, sounding alarm bells among privacy and civil liberties experts. There are also concerns about the 80% accuracy threshold, which critics say is arbitrary and too low, due to potential consequences for those marked as matches. Of India lack of comprehensive data protection laws make the matter more interesting.
The documents add that even if a match is less than 80 percent, it will be considered a “false positive” rather than a negative, which would subject the individual to “full verification with other corroborating evidence.”
“This means that while facial recognition is not giving them the results they themselves have decided to be the threshold, they will continue to investigate,” said Anushka Jain, surveillance policy consultant associate. and technology with the IFF, who submitted this information, said. “This can lead to personal harassment just because the technology says they look like the person the police are looking for.” She added that this move by the Delhi Police could also lead to harassment of people from communities that have been targeted by law enforcement officials.
In response to the IFF’s request for records, police said they were using crime photos and profile photos to run facial recognition. They added that these could be used as evidence but declined to share further details. However, they clarified that in the event of a positive outcome, police officials would conduct further “empirical investigation” before taking any legal action. Delhi Police did not respond to WIRED’s email request for comment.
Divij Joshi, who has spent time researching the legitimacy of facial recognition systems, says the 80% matching threshold is mostly meaningless. Joshi explains that exact numbers are highly dependent on the conditions for testing facial recognition technology models against specific benchmark data sets.
Joshi, a doctoral candidate at University College London, said: “The normal accuracy of a face recognition or machine learning system was determined by comparing a model developed on training data and validation data against the benchmark dataset. “Once the training data is edited, it has to be compared to a third-party dataset or a slightly different dataset.” According to him, this benchmark is often used to calculate the percentage of prediction accuracy.
Evidence of racial bias in facial recognition models has has long made the use of technology problematic. And while many variables affect the accuracy of facial recognition systems, widespread police use of a system with an overall accuracy threshold of 80% seems highly unusual. One Research by the US National Institute of Standards and Technology in 2021 found that the systems used to match a traveler’s face scan to a database of their photos had an accuracy rate of 99.5% or better. Other studies, however, have found error rates as high as 34.7 percent when used to identify women with darker skin tones.
