Towards a writer-dependent hand-written character recogniser

Abstract

The pre-processing of character images prior to character classification is a crucial step in the design of a reliable handwriting recognition system. However, the character structure must be preserved. A preprocessing algorithm is presented and applied to a writer-dependent hand-written character recognition system. The pre-processing stage provides a medial axis polygonal representation of each character which preserves the character structure, and efficient data reduction, without introducing artifacts such as false limbs, "necking" and spurs. This clean character representation allows dynamic information to be recovered. A dynamic feature extractor and a statistical classifier based on dynamic component warping is described. Recognition rates of 91.67% (first choice) and 94.55% (second choice) have been achieved.