Abstract:

The recognition of unconstrained handwritten offline Chinese text line is a difficult problem in current character recognition domain.In order to reduce the interferences from negative samples during text line recognition,a probabilistic model is developed.In this model,negative samples are treated as an information source and are integrated with other kinds of information such as the recognition information of isolated characters and the geometric information of text line.By using only two classifiers,the proposed probabilistic model can be successfully implemented.Experimental results on a real multiple-writer handwritten text database show that,by using the proposed method,the correct recognition rates reach 61.29% without any language model and 72.73% with a bi-gram language model,respectively,which means that the method is effective.

Key words: character recognition, Chinese text line recognition, multiple information fusion, probabilistic model, negative sample