基于胶囊网络与深度置信网络融合模型的手写汉字识别

Handwritten Chinese Character Recognition based on the Fusion Model of Capsule Network and Deep Belief Network

针对离线手写汉字的特征提取困难、不能准确识别等问题,提出了一种胶囊网络与深度置信网络的融合模型。首先从CASIA-HWDB1数据集中随机选择了一些文本分别训练胶囊网络和深度置信网络,然后采用胶囊网络和深度置信网络的融合策略进行了手写汉字识别实验。实验结果表明,在不确定方向上使用汉字融合模型的错误率降低了5.2%,与单独使用胶囊网络和深度置信网络相比,具有更好的识别效果。

Aiming at the difficulties in feature extraction of offline handwritten Chinese characters and inaccurate recognition,this paper proposes a fusion model of Capsule Network(CapsNet)and Deep Belief Network(DBN).First,some texts from CASIA-HWDB1 data set are randomly selected to train CapsNet and DBN respectively.Then,handwritten Chinese character recognition experiments are conducted using the fusion strategy of CapsNet and DBN.The experimental results show that the error rate of fusion model is reduced by 5.2%for Chinese characters in uncertain direction,and it has better recognition effect than using CapsNet and DBN alone.