基于胶囊网络的汉字笔迹鉴定算法

由于采集脱机汉字手写样本时忽略了书写人的心理和生理等因素对书写活动的影响，因而传统笔迹鉴定算法的泛化能力较低。针对上述问题，提出基于胶囊网络的汉字笔迹鉴定算法，并构建了跟踪采集数据集以模拟复杂背景下产生的汉字。胶囊网络构建活动向量表示特定类型的实例化参数，通过动态路由算法将活动向量路由到下一层相应的胶囊中，使下一层胶囊得到更清晰的输入信号。分别采用5种算法对HWDB1.1数据集和跟踪采集数据集进行了测试，实验结果表明：本文算法的分类准确率比其他4种算法的都高，HWDB1.1数据集和跟踪采集数据集中算法的分类准确率分别为95.82%, 94.39%；本文算法具有较强的泛化性能，对训练样本数的依赖程度较低，弥补了卷积神经网络池化层的信息丢失缺陷。

Research on Chinese Character Handwriting Identification Based on Capsule Network

Since the influence of the psychological and physiological factors of the writer on the writing activity was neglected while collecting the offline Chinese handwritten samples, the generalization ability of the traditional handwriting identification algorithm was low. A Chinese character handwriting identification algorithm based on capsule network was proposed, and a complex background of tracking the collected datasets to simulate Chinese characters was constructed. The capsule network constructed an activity vector to represent a specific type of instantiation parameter. The dynamic routing algorithm routed the activity vector to the corresponding capsule in the next layer to enable the next layer capsule to get a clearer input signal. The experimental results of five algorithms in HWDB dataset and tracking acquisition dataset showed that the classification accuracy of this algorithm was higher than that of the other four algorithms. The classification accuracy of HWDB dataset and tracking dataset algorithm were respectively 95.82% and 94.39%. The algorithm had strong generalization performance and low dependence on the number of training samples, making up for the convolutional neural network pooling layer information lost.