A new scheme for rectifying recognition results of printed chinese characters

摘要

Chinese character recognition error occurs when the correct character is not the highest rank character in a candidate set or, more seriously, is excluded from the set. This paper presents a new scheme aimed at rectifying the above error in printed Chinese character recognition. The rectification scheme employs a line-segment cancellation method with low sensitivity to misshapen line segments caused by noise and hence powerful discriminating ability to inspect character images. In addition, the scheme can determine possibly invalid candidate sets, which may exclude the correct characters, and then, for the invalid sets, retrieve additional possibly correct characters to be inspected with the cancellation method. Therefore, the scheme is not only able to pick out the correct character not only in the highest rank position but can also recover the correct character originally excluded from a candidate set. Experimental results showed that 75% of the recognition errors for normal quality documents and 48% for low quality ones were rectified by our new rectification scheme.