A Comparative Study of Learning-based Approaches for Chinese Character Recognition

This paper presents a comprehensive comparison study of various learning-based approaches for Chinese Character Recognition (CCR). The study examines eight types of models that belong to the machine learning model and deep learning model categories. These models include Bagging k-Nearest Neighbor, Random Forest, Support Vector Machine (SVM), Bagging Decision Tree, Xception, LeNet, Multi-Layer Perceptron (MLP), and Visual Geometry Group 16 (VGG16). To conduct the study, a dataset of handwritten Chinese characters is collected. The dataset consists of 5,000 samples distributed across 10 classes of Chinese characters. From the experiment results, we conclude that the best-performing algorithm for the classification model is VGG16, which achieved the highest accuracy score among the eight learning-based models tested in the study. Specifically, VGG16 scored a remarkable accuracy of 99.20%, outperforming the other seven learning-based models. These findings demonstrate the potential of deep learning models, such as VGG16, to improve Chinese character recognition algorithms and enhance their accuracy and performance.