Dynamic Handwriting Recognition of Chinese Characters Based on Pressure Sensor Array

• Main Authors: Kong, Yongkang, Meng, Wujun, Cheng, Guanyin, Zhao, Fubang, Yang, Fuping, Wei, Dapeng
• Format: Conference Proceeding
• Language: English
• Subjects: attention mechanism; Biometrics (access control); Handwriting recognition; Heuristic algorithms; multi-scale residual; pressure sensor array; Pressure sensors; Stroke (medical condition); Three-dimensional displays; time series; Writing
• ISSN: 2689-6621
• DOI: 10.1109/IAEAC59436.2024.10503945

Biometrics have natural uniqueness and reliability, among which handwriting is widely used for personal identification due to its non-invasive and easy-to-capture characteristics. The intricate nature of Chinese characters, a pivotal communication medium for the Chinese, presents significant recognition challenges. Predominantly, existing methods in Chinese character and handwriting recognition rely on static binarized images for identification and authentication, often overlooking individual differences in stroke order and pressure. This research proposes a novel approach for simultaneous recognition of handwritten Chinese characters and handwriting via an array of pressure sensors, which are capable of discerning variations in stroke pressure among different individuals. Moreover, a 3D convolutional neural network with multi-scale residual combined with attentional mechanism (MSR-A-Net) was designed by dynamically capturing time-series pressure data to capture the stroke order of writing. The approach achieved a 97.25% accuracy rate on a custom-compiled dataset featuring ten Chinese characters and four distinct handwriting styles. These results show the potential in the application on Biosecurity detection and human-computer interaction.