How facial recognition sees you, matches you, and sometimes gets it wrong A face in a crowd. A camera. An algorithm. And, potentially, a police alert. Facial recognition technology used by Delhi Police is back under scrutiny after Opposition leaders questioned its deployment during recent Cockroach Janta Party (CJP)-led protests at Jantar Mantar, amid media reports highlighting possible fault lines in FRT and raising questions over police claims. Delhi Police maintains that its Facial Recognition System does not establish identity on its own: a possible match generates an alert, after which “investigators are expected to independently verify the person.” That distinction is central to understanding the technology. A machine does not “recognise” a face the way a human does. It converts facial features into a numerical representation and calculates how closely it matches another image. The controversy therefore raises two questions: how reliable is the technology, and how much should police be trusted with a system capable of identifying people in public spaces? India Today’s Open Source Intelligence (OSINT) team examined research papers, technical studies and other open-source material to understand how facial recognition works, where it can go wrong and the still-unsettled legal framework governing its use in India. So how does the machine see you? The starting point is deceptively simple. The Journal of Law and Technology at Texas explains that an image stored in a computer is essentially an "array of numbers". The software first has to locate a face within that numerical grid before it can attempt to recognise it. Older geometric systems did this by measuring facial landmarks, including the relative positions of the eyes and nose and the size and shape of particular features. Photometric techniques approached the problem differently. Rather than relying only on facial landmarks, they analysed patterns across the entire image.