Configuring Hand-Writing Recognition Systems from Form Knowledge

Effective handwriting recognition (HR) has always been an important but distant goal for Artificial Intelligence (AI) research. This is primarily due to the complexity of the task – a truly effective HR system must cope with a vast diversity of writing and character styles as well as a range of writing conditions. The traditional approach has been to build highly specialised systems capable of coping with a narrow type of HR task (eg, postal addresses [4,5], tax-form data [3], or bank cheques [1,2]). Recently, researchers have had some success in developing more general HR systems, but these systems suffer from greatly reduced recognition reliability, and ultimately depend on a significant training time to tune their recognition processes to the writing style of a particular individual.