India trials biometric boarding; new study evaluates facial recognition accuracy for Indian faces
Six Indian airports are trialing facial recognition technology for biometric boarding while Indian researchers have re-run large-scale tests that proved highly controversial when they revealed the technology’s failings with darker skin tones, but this time on Indian faces. The results could put into question the viability of biometric boarding in airports.
The boarding trials at airports including Bengaluru, Hyderabad and Pune, reported by Business Today, are expected to be successful and the technology subsequently applied to other airports.
India’s aviation minister has the Airports Authority of India plans to invest R250 billion ($3.4 billion) over the next four to five years for expanding and developing terminals using modern technology.
Evaluating facial recognition accuracy on Indian faces
But new tests of facial recognition software on datasets of photos of Indian faces show that the gender of Indian women is inaccurately identified 7 percent to the time, reports the Scroll.
Back in 2018 when the scandal of racial biases in facial recognition algorithms broke, Smriti Parsheera, a lead technology policy researcher at the National Institute of Public Finance and Policy in Delhi at the time, realized that the datasets being used were for predominantly all Black or all White. She knew that the technology still had to be tested on Indian datasets.
Pahseera worked with technology policy researcher Gaurav Jain to test Amazon’s Rekognition, Microsoft Azure’s Face, Face++ and India’s FaceX (based in Bengaluru whose airport is included in the facial recognition boarding trial) on the public database of election candidates found on the Election Commission of India website.
Algorithms have been constantly updated since the 2018 US study, but the pair found concerning results. Azure failed to detect the largest number of Indian faces at a little over 1,000 giving it an error rate just over three percent.
India’s FaceX failed to detect more than 800 Indian faces, giving a 2.6 percent error rate. But it also misidentified the gender of Indian women almost 11 percent of the time compared to 1.35 percent of the time for Indian men.
“We thought that FaceX might be better because it’s an Indian company so they should have clients in India and use Indian faces, but their results were quite surprising,” Scroll quotes Jain as saying. This is of particular concern to technology policy lawyer Divij Joshi who says that most Indian government contracts for facial recognition go to Indian firms. The government is currently tendering for facial recognition equipment for exam centers and recently authorized the automated roll out of biometric facial recognition in police investigations.
China’s Face++ brought its error rate for Black women down from 35 to 4 percent since 2018 but sill misidentified Indian women 15 percent of the time.