Financial Times, Creative Business Supplement 18/03/03
Machines that see more clearly
By Alan Cane
Reproduced by kind permission of FT.com
Dr Who's Daleks managed with an inverted sink plunger, but most
machine-vision systems struggle to "see" the real world through humanoid
eyes. The problem is that the processes by which the brain interprets
visual stimuli are not completely understood, so it is hard to create the
electronic equivalent.Which does not, of course, prevent engineers
The Hungary-based company AnaLogic has been developing processors
which it likens to the eye's retina. Now a Cambridge-based start-up,
break-step productions, has a software solution which its managing
director, Patrick Andrews, says is based on neurophysiology.
The system can recognise a broad range of objects as opposed to, say,
only text or industrial components.
The system, called Foveola, can interpret the
edges of an object under scrutiny so that a unique mathematical signature
can be generated. It can only recognise things it has seen before, but
its memory is cumulative - once seen not forgotten.
Because of its small
size and low cost, Andrews believes Foveola has immediate application in
educational toys and interactive robots, as well as in brand protection
and detecting forged company logostyles on the internet.