Chris Williams, a post-graduate researcher at Aston University, has been working on a project to recognise the US Zip codes, the equivalent of post codes, on envelopes for the US Postal Service. The project has taken a slightly different approach to handwritten character recognition and is looking to exploit the contextual information that exists in handwritten strings of data. So, for example, if there are two number sevens written in a string of digits, then there is a good chance that the second one will be written in a similar style to the first, and this prior information can be used to help recognise the next number seven. The recognition approach is to use ink generators that create circles that shrink around the digit being recognised to give the correct identification. In effect, what the system does is splat some hypothetical ink at the digit to be recognised. Each blob of ink centres around the digit if it lands on a part of it, and then that blob shrinks to fit the digit being recognised. The problem with this system so far is that it takes about one second to recognise each digit, not yet fast enough for commercial use.
This article is from the CBROnline archive: some formatting and images may not be present.
CBR Online legacy content.