Using Shape models for Error Correction of On-line Handwriting

A writer-dependent technique is proposed for automatically correcting written spelling errors in on-line cursive handwriting. In a training process, handwriting samples of the user are collected and segmented into characters by using a writer-independent segmentation algorithm. Shape models are built on these samples to capture the user’s writing style for each character. In the application process, handwriting errors are accurately located using a shape model-based character spotting algorithm. The handwriting trajectory is segmented at the error location and if necessary, a new character is generated from shape models and connected with previous strokes smoothly with the guidance of a conditional sampling algorithm. Experimental results show that the corrected handwriting still maintains the user’s writing style without perceptible artifacts.