Record-setting facial recognition software tests near perfect
Thursday, July 24, 2014 - 02:10
Computer scientists from Hong Kong University have unveiled the most accurate facial recognition software in the world. The algorithm, which scored a near perfect 99.15 per cent accuracy rating in tests, could prove to be a game-changing technology in automated crowd security. Ben Gruber reports.
▲ Hide Transcript
▶ View Transcript
STORY: Surveillance cameras have proved a powerful tool when monitoring large crowds for signs of danger. But pinpointing a suspicious person or activity can be like finding a needle in a haystack.
Soon that job may prove easier, with the help of a new facial recognition computer program developed by professor Xiaoou Tang and his team at Hong Kong University.
(SOUNDBITE) (English) CHINESE UNIVERSITY OF HONG KONG, FACULTY OF ENGINEERING, PROFESSOR TANG, SAYING:
"A couple of months ago we got a result that actually surpassed the human performance in recognising faces."
Humans recognize faces at an accuracy rating of 97.53 percent. Tang's software scored a near perfect 99.15 percent, besting humans and all other facial recognition programs to date.
Tang's program was tested using thousands of picture sets to show how accurately it can recognise whether two photographed faces were of the same person, regardless of changes in lighting, make-up and camera angles.
Assistant professor Xiaogang Wang says this software could help law enforcement and security agencies to seek out individuals among a crowd of thousands. And by interpreting facial expressions, it could predict violent activity before it starts.
(SOUNDBITE) (English) ASSISTANT PROFESSOR, DEPARTMENT OF ELECTRONIC ENGINEERING, XIAOGANG WANG, SAYING:
"So we try to develop a computation model try to detect and check and analyse the behaviour of the crowd and predict the potential crowd disasters. But the traditional way, the video surveillance, they only focus on a small number of objects in a very simple environment. But now what we are doing is we target thousands of objects in very complex environments. So this problem is very challenging and we are working in that direction."
It's a problem that Wang and Tang believe can be solved as computer processing speeds increase with time. For now they're content knowing their algorithm could lay the foundation for the next generation of security systems that will keep people safe and ultimately save lives.
Press CTRL+C (Windows), CMD+C (Mac), or long-press the URL below on your mobile device to copy the code