Polity17 October 20263 min read
Supreme Court says police use of facial recognition must meet proportionality test
The Supreme Court has said it will judge the deployment of facial recognition technology (FRT) by the test of proportionality. It has also directed that law enforcement strictly tie real-time use of the technology to existing criminal data. The observation comes amid a wider debate on FRT and surveillance evasion, which gained traction after suggestions that creative makeup could help people evade facial recognition cameras during protests.
What the Supreme Court said
The Court said the principle of proportionality, established in the K.S. Puttaswamy (Right to Privacy) judgment of 2017, will be the touchstone for evaluating FRT deployment. It directed that real-time use by law enforcement be tied to existing criminal data, a condition meant to curb broader monitoring of crowds.
On whether police can scan crowds and protests, the answer so far is that they can use the technology only within the bounds of proportionality and the criminal data restriction. The Court's observations do not amount to a statute, and that is where the gap lies.
How the technology works
FRT is a biometric system that identifies or authenticates people by mapping their facial features from digital images or video. Modern systems use Artificial Intelligence to extract distinguishing patterns instead of comparing images pixel by pixel. They convert visual traits into a numerical representation known as a face embedding or template.
There are two broad uses. In 1:1 verification, a face is checked against a single stored identity, as when unlocking a smartphone or verifying a traveller at an airport gate through Digi Yatra. In 1:n identification, a system scans crowds through CCTV and looks for matches in a criminal database. Law enforcement relies on the second use.
A match depends on similarity scores and matching thresholds. A positive match is therefore not conclusive proof of identity and requires independent verification.
Attempts to evade it
The makeup idea has precedents. In 2010, the artist Adam Harvey explored, through the CV Dazzle Project, whether high-contrast, bold makeup and asymmetrical hairstyles could break automated pattern recognition. The project drew on the naval "Dazzle" camouflage of World War I, which used geometric shapes to confuse enemy targeting instead of hiding ships.
A 2021 Ben-Gurion University study went further. It showed that machine-learning-generated makeup patterns, aimed at highly identifiable facial regions, could bypass FRT software with up to 98% success.
The legal gap and the concerns
The Digital Personal Data Protection (DPDP) Act, 2023 regulates personal data. India, however, lacks a dedicated statutory framework for police FRT deployment, particularly on watchlists, matching thresholds, data retention and independent audits.
The technology also raises problems of accuracy and rights. Algorithms are prone to false positives, particularly along lines of race and gender, which can lead to wrongful arrests and bias against marginalised communities. The fear of being identified, profiled and added to state databases can also deter citizens from exercising the fundamental right to peaceful assembly under Article 19.