Contextual post-processing based on the confusion matrix in offline handwritten Chinese script recognition

The inclusion of potentially correct characters in candidate sets is key to improving accuracy in the recognition of Chinese scripts in the aspect of contextual post-processing. This paper presents two methods based on a confusion matrix to recall the correct characters. The first method uses original candidates to conjecture the most likely correct characters, and then combines the conjectured set with the original candidates to produce a new candidate set. The second method performs an approximate matching of adjoining characters in a sentence with Chinese words so as to recall the most likely correct character. Experimental results demonstrate the effectiveness of our proposed methods.