Abstract:

As to the low accuracy of offline handwritten Chinese character recognition, a cascade classifier model based on Modified Quadratic Discriminant Function（MQDF） and Deep Boltzmann Machine（DBM） is proposed. The main idea behind the model is that MQDF and DBM can complement each other in features extraction and classification mechanisms. Firstly it recognitzes and gets the result using MQDF, and calculates the generalized confidence of the recognition result. If the confidence meets the requirement, the recognition result can be as the final result directly output; otherwise, combining with DBM to make recognition again, and the result will be taken as the final recognition result. Experimental results show that using MQDF-DBM model can achieve higher accuracy than using MQDF and DBM model alone, and the recognition speed is faster than DBM.

Key words: Modified Quadratic Discriminant Function（MQDF）, Deep Boltzmann Machine（DBM）, offline handwritten Chinese character recognition

摘要：

针对脱机手写体汉字识别准确率较低的问题，提出一种基于修正的二次判别函数（Modified Quadratic Discriminant Function，MQDF）与深度玻尔兹曼机（Deep Boltzmann Machine，DBM）的分类器级联模型。该模型的主要思想是MQDF和DBM在特征提取和分类机制上可以相辅相成。先用MQDF进行识别并得出结果，同时计算该结果的一个广义置信度。若置信度满足要求，则将识别结果作为最终结果输出，否则结合DBM进行二次识别，得到最终识别结果。实验结果表明，使用MQDF-DBM模型可以获得比单独使用MQDF和DBM模型更高的识别准确率，且识别速度比DBM更快。

关键词: 修正的二次判别函数, 深度玻尔兹曼机, 脱机手写汉字识别