Online Handwriting Recognition

Online handwriting recognition studies how to convert the sequence of pen positions, strokes, and pen-up or pen-down events into text. Systems combine stroke features with recurrent or other sequence models, and use techniques such as CTC decoding, subword tokenization, and self-supervised pretraining; sensor-based systems may also learn from inertial signals. Evaluation spans character and word recognition, often testing whether models generalize to new writers, scripts, or distorted strokes. A central challenge is balancing accuracy and speed while handling writer variation and scarce training data, especially across languages and writing systems.