Offline handwritten Chinese character recognition based on DBN fusion model

Abstract

For the past several decades, offline handwritten character recognition is widely and deeply studied. However, the requirements of the identification results are constantly improving in practical applications. This paper presents a new classifier cascaded model to improve the accuracy of offline handwritten Chinese character recognition. New model is the fusion of modified quadratic discriminant function (MQDF) and deep belief network (DBN). The main idea behind MQDF-DBN fusion model is that the significant difference on features and classification mechanisms between MQDF and DBN can complete each other. First to recognize and get result using MQDF, calculate the recognition confidence as evaluation criteria. If recognition confidence is high, the recognition result of MQDF will be output directly. Otherwise, using the DBN to make recognition again and getting the final recognition result. Experiment shows that the fusion model of MQDF and DBN proposed in this paper has achieved better accuracy than the single use of MQDF and DBN in the offline handwritten Chinese character recognition task, which is performed on the ETL-9B handwritten Chinese character dataset.