A structural method with local refining for handwritten character recognition

A new algorithm for the recognition of the numerical amount on bankchecks is presented. Mainly, it results from the modification of the well known Baptista-Kulkarni algorithm (1988). The improvements proposed are based on the recognition by parts approach, which promises a better than human character recognition ability in a context free situation. It is shown that with this algorithm a misrecognition rate of lower than 1% can be reached with a reasonably high rejection rate, using numerals extracted from a common database. Better performances could be obtained using numerals extracted from Italian bankchecks, due to the greater attention that people pay to writing amounts on them