Perception of multi-author handprinted text

Handprinted character recognition has a very diverse range of applications which include direct data input devices to replace the ubiquitous keyboard, and reading aids for the blind. The major problems confronting designers of such machines is finding solutions to the recognition of an infinite number of character shapes experienced in normal unconstrained handprinting and its associated variability. Both of these two factors even cause humans to misread handprinted characters when read in isolation. Humans are in general able to correctly read handprinted characters when they are presented in context, and also improve their reading performance with experience. A method to emulate human reading performance is proposed in the paper which can be applied to applications that are inherently interactive such as those previously mentioned.<
>