KEYBOARD skills could die out in the next century if a novel method for processing visual information takes off. Now being developed by British-based researchers, the technique enables computers to lip read as well as recognise voices鈥攋ust like HAL in 2001:A Space Odyssey.
According to Andrew Bangham of the School of Information Systems at the University of East Anglia, who is leading the research, the system succeeds where older technology fails because it maintains the high resolution of the image it is viewing while focusing on a specific area of interest.
Other image recognition technology relies on a technique known as filtering. By removing successive levels of detail, the objects in the image are simplified until they appear as basic shapes that the computer can recognise. The problem is that the picture becomes fuzzier as more information is removed, losing the very detail the computer needs for recognition.
Advertisement
鈥淧erhaps it鈥檚 because we came late on to the scene that we have stumbled on our different technique,鈥� says Bangham. His process, which is called 鈥渟ieving鈥�, sifts through the information in the image looking for shapes of specific sizes.
The team鈥檚 computer is fitted with a microphone as well as a camera. The camera watches the mouth and grades the lip patterns associated with words according to size. 鈥淭he O shape of the open mouth is obviously the largest shape and the closed lips at the end of the word the smallest,鈥� says Bangham.
Having 鈥渟ieved鈥� the shapes, the computer can match them against the sounds the person is making to work out the words. This dual recording overcomes the main problem plaguing purely audio systems鈥攃onfusion arising from background noise. Recognition is distinctly higher than with either audio or video systems, regardless of noise.
The new system already recognises the letters of the alphabet and simple words such as 鈥測es鈥� and 鈥渘o鈥� when spoken by 10 different people. 鈥淲e expect this to be useful within a year at research level,鈥� Bangham says.
