Studies on recognition of on-line writing have mostly been restricted to only a single language, such as either Hangul or English. Since, however, we often write Hangul and English in mixture, needed is a recognizer for language-mixed handwriting.
In this thesis, proposed is a recognizer called 'Unified Recognizer' for Hangul / English / Digit mixed characters. Models of each language were constructed using Hidden Markov Models (HMMs) and connected in parallel, resulting in Unified Network. The recognition result is obtained by searching the statistically optimal path for a given input using Viterbi algorithm.
For language-mixed word recognition, Unified Network has been extended to Circular Unified Network by adding backward arcs from the final node of Unified Network to the initial node. These circle-making arcs model ligatures between characters. Recognition here is similarly defined as the problem of searching the optimal circular path, and is also solved by Viterbi algorithm. This approach can be easily extended to recognizing other alphabetic languages.
