Recognition of broken and noisy handwritten characters using statistical methods based on a broken-character-mending algorithm

A broken-character-mending algorithm is proposed to deal with broken characters. The algorithm can mend various kinds of broken characters if the number of missing pixels is smaller than the number of existing pixels in the mending masks. It also has no negative effects on the pixels that do not need mending. Mending masks of different sizes can be combined and the same mask can be applied several times to improve the mending result. The algorithm has been tested on some seriously broken characters, in which up to 50% of pixels are deleted randomly. The recognition process is based on statistical analysis combined with a shape recognition method. In our experiment, the recognition rate of the statistical method increased by about 7% after using the character-mending algorithm. The substitution rate decreased from 7.4 to 2.3% and the reliability increased from 91.9 to 97.6% after using the broken-character-mending algorithm. © 1997 Society of Photo-optical Instrumentation Engineers.