Cursive character prediction using a deep learning model

Cursive character (CC) recognition faces challenges owing to multifarious CC writing styles and complex character structures. Therefore, a universal CC identification system is essential. This study focuses on Chinese cursive characters and proposes a novel deep learning-based CC prediction model, along with a corresponding mobile application for real-time character recognition. The proposed model comprises three primary steps: data collection (from historical manuscripts, cleaned to remove noise); image preprocessing (including normalization, resizing, and augmentation); and prediction (involving hyperparameter optimization, regularization, and prediction using deep learning models). Transfer learning from convolutional neural networks such as Residual neural network (ResNet), Densely Connected Convolutional Networks (DenseNet), very deep convolutional network (Visual Geometry Group, VGG), and EfficientNet is employed. The model was trained and evaluated on a curated dataset of 99,296 images across 331 character classes, derived from Wang Xizhi’s ‘Grass Jue Song and Cursive Thousand Characters’. Using DenseNet-201, the proposed model achieved superior performance, with accuracy, precision, recall, and F1-score of 0.92, 0.92, 0.93, and 0.92, respectively. In contrast to existing work, which often focuses on smaller datasets or different scripts, this study introduces a large-scale recognition framework tailored for complex and visually similar Chinese cursive characters and demonstrates its practical deployment through a mobile application.