On-line Handwriting Recognition using Hidden Markov Models
An application was developed using C++ on Unix for a handwriting recognition system giving emphasis to pre-processing and recognition stages. Noise filtering and character segmentation were tasks of the pre-processor. Characters, represented as templates in the form of directed strokes were fed to the recogniser containing a database of characters HMMs. The recognition involved training the HMMs using the Baum-Welch algorithm and the classification using the Forward, Forward-Backward and Viterbi algorithms.